Neuroradiolog y/Head and Neck Imaging • Original Research Lu et al. Histogram Analysis of ADC Maps
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
Neuroradiology/Head and Neck Imaging Original Research
Histogram Analysis of Apparent Diffusion Coefficient Maps for Differentiating Primary CNS Lymphomas From Tumefactive Demyelinating Lesions Shan Shan Lu1,2 Sang Joon Kim1 Namkug Kim1 Ho Sung Kim1 Choong Gon Choi1 Young Min Lim 3 Lu SS, Kim SJ, Kim N, Kim HS, Choi CG, Lim YM
Keywords: apparent diffusion coefficient (ADC), demyelinating disease, diffusion MRI, histogram analysis, primary CNS lymphoma (PCNSL) DOI:10.2214/AJR.14.12677 Received February 10, 2014; accepted after revision June 11, 2014. 1 Department of Radiology and Research Institute of Radiology, Asan Medical Center, University of Ulsan College of Medicine, 88 Olympic-ro 43-gil, Songpa-gu, Seoul 138-736, Korea. Address correspondence to S. J. Kim ([email protected]). 2
Department of Radiology, The First Affiliated Hospital of Nanjing Medical University, Nanjing, Jiangsu Province, China.
3
Department of Neurology, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Korea.
AJR 2015; 204:827–834 0361–803X/15/2044–827 © American Roentgen Ray Society
OBJECTIVE. This study intended to investigate the usefulness of histogram analysis of apparent diffusion coefficient (ADC) maps for discriminating primary CNS lymphomas (PCNSLs), especially atypical PCNSLs, from tumefactive demyelinating lesions (TDLs). MATERIALS AND METHODS. Forty-seven patients with PCNSLs and 18 with TDLs were enrolled in our study. Hyperintense lesions seen on T2-weighted images were defined as ROIs after ADC maps were registered to the corresponding T2-weighted image. ADC histograms were calculated from the ROIs containing the entire lesion on every section and on a voxel-by-voxel basis. The ADC histogram parameters were compared among all PCNSLs and TDLs as well as between the subgroup of atypical PCNSLs and TDLs. ROC curves were constructed to evaluate the diagnostic performance of the histogram parameters and to determine the optimum thresholds. RESULTS. The differences between the PCNSLs and TDLs were found in the minimum ADC values (ADCmin) and in the 5th and 10th percentiles (ADC5% and ADC10%) of the cumulative ADC histograms. However, no statistical significance was found in the mean ADC value or in the ADC value concerning the mode, kurtosis, and skewness. The ADCmin, ADC5%, and ADC10% were also lower in atypical PCNSLs than in TDLs. ADCmin was the best indicator for discriminating atypical PCNSLs from TDLs, with a threshold of 556 × 10 −6 mm2 /s (sensitivity, 81.3 %; specificity, 88.9%). CONCLUSION. Histogram analysis of ADC maps may help to discriminate PCNSLs from TDLs and may be particularly useful in differentiating atypical PCNSLs from TDLs.
N
ow frequently encountered in both immunocompetent and immunocompromised patients, primary CNS lymphomas (PCNSLs) constitute from 1% to 6% of primary brain neoplasms [1, 2]. In patients with normal immunity, PCNSL usually presents as a well-demarcated solitary lesion without necrosis or hemorrhage and with marked homogeneous enhancement, both of which are considered to be typical imaging features [3– 5]. However, many atypical radiologic presentations have been reported, such as necrosis, hemorrhage, variably heterogeneous enhancement, and sometimes even no enhancement or as a diffuse lesion without mass formation [6–9]. Correct diagnosis of typical PCNSLs is usually not difficult in routine clinical practice. However, when unusual MRI manifestations are present, differentiating atypical PCNSLs from tumefactive demyelinating lesions (TDLs) can sometimes be challenging,
both clinically and radiologically. The incidence of TDL is estimated as 0.3 per 100,000 per year [10]. They can cause symptoms mimicking brain tumors and commonly have heterogeneous enhancement [11]. However, accurate recognition of PCNSL according to the imaging criteria is important to avoid inappropriate use of steroid medication and to ensure timely and adequate treatment. Many advanced imaging techniques have been used for distinguishing between PCNSLs and TDLs. Choline/creatine (Cho/ Cr) ratios and choline/N-acetyl aspartate (Cho/NAA) ratios derived from MR spectroscopy may differ between the two diseases, although there are still some overlaps [2, 12]. Similarly, in dynamic susceptibility contrast T2*-weighted perfusion MRI, PCNSLs tend to have higher relative cerebral blood volume than TDLs, although sometimes there is only a slight difference [13]. The role of diffusion-weighted imaging (DWI) with apparent diffusion coeffi-
AJR:204, April 2015 827
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
Lu et al. cient (ADC) maps is currently being actively investigated. PCNSL has been reported to have a tendency toward low ADC because of its high cellularity, which helps to differentiate it from toxoplasmosis or gliomas [14– 16]. However, to our knowledge, the role of DWI in distinguishing PCNSLs from TDLs has not yet been reported, and there are only a limited number of reports evaluating ADC in the subgroup of atypical PCNSLs. To measure ADC in brain lesions, most of the previous studies used ROI analysis, which was based only on the representative part of the tumor [17–19]. However, selecting a location is subjective and may be difficult when there are atypical imaging features. Recently, ADC histogram analysis has been found to be useful for grading gliomas and predicting the treatment response and progression-free survival in patients with glioblastomas [20, 21]. Given that histogram analysis includes an entire lesion, it may reveal the heterogeneity of lesions and thus ensure more accurate discrimination. In our study, we intended to evaluate the clinical utility of histogram analysis of ADC maps for differentiating PCNSLs, especially atypical PCNSLs, from TDLs, and we further investigated the diagnostic performance of ADC histograms for this discrimination.
regression of the lesions and clinical symptoms during follow-up after corticosteroid treatment. The final diagnosis of TDL was made on the basis of either the histopathology findings or a strong clinical suspicion. Among the TDLs, five patients were confirmed by stereotactic biopsy and 13 were diagnosed by clinical follow-up and repeat MRI. PCNSL patients were further divided into two subgroups on the basis of conventional MRI: typical PCNSLs had a nodular pattern with intense homogeneous enhancement and a well-defined margin accompanied by surrounding edema (n = 31), and atypical PCNSLs showed necrosis, hemorrhage, or atypical enhancing patterns on contrast-enhanced T1-weighted imaging, including patchy infiltrative enhancement, streaky infiltrative enhancement, ring enhancement, or even lack of enhancement (n = 16) [24].
Image Acquisition All of the images were obtained on a 3-T MR system (Achieva, Philips Healthcare) with an eight-channel head coil. The conventional protocols included axial T1-weighted (TR/TE, 500/10; slice thickness, 5 mm; slice gap, 2 mm; FOV, 230 × 183 mm; matrix, 256 × 183; number of signals acquired, 1; and voxel resolution, 0.90 × 1.00 × 5.00 mm), T2-weighted fast spinecho (TR/TE, 3000/80; slice thickness, 5 mm; slice gap, 2 mm; FOV, 230 × 184 mm; matrix, 400 × 255; number of signals acquired, 1; and voxel resolution, 0.58 × 0.72 × 5.00 mm), FLAIR (TR/TE, 10,000/125; slice thickness, 5 mm; slice gap, 2 mm; inversion time, 2200 ms; FOV, 230 × 182 mm; matrix, 352 × 230; number of signals acquired, 1; and voxel resolution, 0.65 × 0.90 × 5.00 mm), and contrast-enhanced T1-weighted 3D fast
Materials and Methods Patients This retrospective study was approved by our institutional review board, and the requirement for informed consent was waived. We reviewed our institutional database for the time period between April 2007 and May 2012 and selected 47 PCNSL patients (33 men and 14 women; age, 29–82 years) and 18 TDL patients (10 men and eight women; age, 22–66 years). Our inclusion criteria were histopathologic diagnosis of PCNSLs by surgical resection or stereotactic biopsy, histopathologically proven or clinically diagnosed TDLs, and raw DWI data available with good quality along with the conventional MRI results. Eighteen PCNSLs and five TDLs were excluded because of inadequate DWI, including incomplete raw data and data with motion artifact or image distortion. The clinical diagnosis of TDLs was based on the following criteria [12, 13, 22, 23]: acute or subacute onset of neurologic symptoms and signs; at least one brain lesion with the longest diameter more than 2 cm on MRI; supportive CSF analysis, including oligoclonal bands, IgG index, myelin basic protein, or supportive somatosensory-evoked potential tests; no evidence of systemic illness, vasculitis, toxic or metabolic disease, or CNS infection; and documented
828
A
B
C
D
Fig. 1—64-year-old man with atypical primary CNS lymphoma. A and B, Multifocal hyperintensity lesions are seen on FLAIR images involving bilateral frontal cortex and subcortical white matter (A) and with patchy infiltrative enhancement in left frontal lobe (B). C and D, Diffusion-weighted image (C) and corresponding apparent diffusion coefficient (ADC) map (D) show obvious diffusion restriction (arrow, D). (Fig. 1 continues on next page)
AJR:204, April 2015
Histogram Analysis of ADC Maps
250 Frequency
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
300
200 150 100 50 0 400 ADCmin
600 800 ADC5% ADC10%
1000
1200
1400
1600
1800
ADC Value (10 mm /s) −6
2
E Fig. 1 (continued)—64-year-old man with atypical primary CNS lymphoma. E, ADC histogram of lesions is asymmetric, with low minimum ADC (ADCmin), 5th-percentile ADC (ADC 5%), and 10th-percentile ADC (ADC10%) of 472, 622, and 648 × 10 −6 mm2 /s, respectively (arrows).
field echo (TR/TE, 9.8/4.6; slice thickness, 1 mm; FOV, 224 cm; matrix, 224 × 224; number of signals acquired, 1; and voxel resolution, 1.0 × 1.0 × 1.0 mm) imaging with transverse, coronal, and sagittal planes after IV injection of 0.1 mmol/kg of gadoterate meglumine (Dotarem, Guerbet). Echo-planar DWI was performed in the axial plane with b = 0 and 1000 s/mm2. Other parameters included TR/TE, 3000/56; slice thickness, 5 mm; slice gap, 2 mm; FOV, 250 × 250 mm; matrix, 128 ×125; number of signals acquired, 1; and voxel resolution, 1.95 × 2.00 × 5.00 mm. DWI was performed in three orthogonal directions of diffusion sensitizing gradients and combined into isotropic images.
er in-plane resolution than that of DWI (1.95 × 2 mm), and therefore ADC maps were coregistered to T2-weighted imaging. Motion correction was performed to correct a rigid body motion to the time series data. This was done automatically in the Nordic ICE software. The high-signal-intensity area on T2-weighted imaging was identified as the outermost lesion margin. ROIs were manually drawn around the entire lesions on all of the axial sections from the coregistered T2-weighted–ADC images with reference to the underlying T2-weighted image. Necrosis, edema, and hemorrhage were all included and the contrast-enhanced portion seen on contrast-enhanced T1-weighted imaging was always included. The ADC values were calculated on a
voxel-by-voxel basis from all of the ROIs and recorded in a separated file using NordicICE software. The files were then transported to an inhouse software developed using Matlab 2010b (MathWorks) in a blinded manner. The histogram of the ADC values was plotted on the x-axis with a predefined total bin number of 500. The y-axis was recorded as the frequency of each bin. To fit the ADC histograms, which are usually distributed broadly and asymmetrically in tumors, we used gaussian mixture modeling in this study. The gaussian mixture number was empirically set as 2 or 3 for generating a suitable histogram fitting curve. All of the gaussian mixture number and fitting curves were checked and verified by a neuroradiologist with 20 years of experience to ensure the best curve fitting. We evaluated the various representative parameters, including the mean ADC value (A DCmean); mode ADC value (ADCmode, the peak height position of the histogram); minimum ADC value (ADCmin); and skewness, kurtosis, and SD of the histograms. We also measured the six cumulative histogram parameters including the 5th (ADC5%), 10th (ADC10%), 25th (ADC25%), 50th (ADC50%), 75th (ADC75%), and 90th (ADC90%) percentiles. The nth percentile was the point at which n% of the voxel values that form the histogram were found to the left [20].
Statistical Analysis All of the data were reported as the mean ± SD. Conventional MRI features of PCNSLs and TDLs were compared using the Pearson chi-square test.
Image Processing and Analysis Conventional MRI features, including the number of lesions, enhancing pattern, hemorrhage, and necrosis, were analyzed. The enhancing patterns were classified into six types: nodular pattern, lesions with well-defined margins and homogeneous enhancement; ring pattern, complete rimlike enhancement in the periphery without enhancement in the center; open-ring pattern, incomplete rimlike enhancement in the periphery; patchy infiltrative pattern, ill-defined or irregularly marginated lesions with heterogeneous enhancement; streaky infiltrative pattern, linear or dotlike enhancement; and nonenhancing pattern, abnormal signal intensity on T2-weighted imaging without definite contrast enhancement. All of the raw DWI data were transferred from the MRI scanner to an independent PC for generating ADC maps and image coregistration using a dedicated software package (NordicICE, NordicNeuroLab). ADC maps were generated using the monoexponential expression for diffusion attenuation after spatial and temporal smoothing. T2-weighted imaging (0.58 × 0.72 mm) had high-
A
B
Fig. 2—29-year-old man with atypical primary CNS lymphoma (PCNSL) mimicking tumefactive demyelinating lesion (TDL). A and B, Confluent hyperintensity lesions are observed on FLAIR image (A) in both frontal periventricular white matter and parietooccipital white matter and with streaky infiltrative enhancement seen on contrast-enhanced T1-weighted image (B). (Fig. 2 continues on next page)
AJR:204, April 2015 829
The ADC histogram parameters were compared using the Student t test in all of the PCNSLs and TDLs and then between atypical PCNSLs and TDLs. ROC curve analyses were performed to evaluate the diagnostic performance of each histogram parameter and to determine the optimum thresholds for discriminating PCNSLs, especially atypical PCNSLs, from TDLs. The AUC, sensitivity, and specificity were further calculated using the ROC curve. All analyses were performed using commercially available statistical software packages (SPSS, version 13.0, IBM, or MedCalc, version 12.3.0, MedCalc Software). The p value was two-sided and a value of less than 0.05 was considered statistically significant.
C
D
350 300 250 Frequency
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
Lu et al.
200 150 100 50 0 500 600 700 800 ADCmin ADC5% ADC10%
900
1000
1100
1200
1300
1400
1500
ADC Value (10 mm /s) −6
2
E Fig. 2 (continued)—29-year-old man with atypical primary CNS lymphoma (PCNSL) mimicking tumefactive demyelinating lesion (TDL). C and D, Diffusion-weighted images show prominent increased signal (C) with partially decreased apparent diffusion coefficient (ADC) in peripheral portion (arrows, D). E, ADC histogram shows low minimum ADC (ADCmin), 5th-percentile ADC (ADC 5%), and 10th-percentile ADC (ADC10%) of 556, 669, and 697 × 10 −6 mm2 /s, respectively, suggesting PCNSL rather than TDL.
TABLE 1: Conventional MRI Features of Primary CNS Lymphomas (PCNSLs) and Tumefactive Demyelinating Lymphomas (TDLs) Feature
PCNSLs (n = 47)
TDLs (n = 18)
p
Single lesion
26 (55.3)
5 (27.8)
0.047
Multiple focal lesions
15 (31.9)
9 (50.0)
0.176
Diffuse lesions
6 (12.8)
4 (22.2)
0.575
31 (66.0)
0 (0.0)
0.000
No. of lesions
Enhancing pattern Nodular Ring
4 (8.5)
1 (5.6)
1.000
Open ring
0 (0.0)
6 (33.3)
0.000
Patchy infiltrative
7 (14.9)
3 (22.2)
1.000
Streaky infiltrative
4 (8.5)
1 (0.0)
1.000
1 (2.1)
7 (38.9)
0.000
Necrosis
No enhancement
5 (10.6)
6 (33.3)
0.070
Hemorrhage
4 (8.5)
1 (5.6)
1.000
Note—Data in parentheses are percentages.
830
Results Visual Assessment of Conventional MRI and DWI On conventional MRI, a single lesion was observed in 26 PCNSL patients and five TDL patients, and multiple lesions were found in 15 PCNSL patients and in nine TDL patients. Diffusely distributed lesions were observed in the remaining six PCNSL patients and four TDL patients. In nine of 16 (56.3%) atypical PCNSL patients, lesions could not be discriminated from edema. The following atypical features were found in a substantial proportion of the atypical PCNSL patients: necrosis (n = 5), hemorrhage (n = 4), ring enhancement (n = 4), patchy infiltrative enhancement (n = 7), streaky infiltrative enhancement (n = 4), or absent contrast enhancement (n = 1). Eleven of 18 TDLs showed varying enhancement, including an open-ring pattern (n = 6), rimlike pattern (n = 1), patchy infiltrative pattern (n = 3), and streaky infiltrative pattern (n = 1). No obvious enhancement was observed in the other seven TDLs. The conventional MRI manifestations of PCNSLs and TDLs are summarized and compared in Table 1. Most TDLs appeared predominantly isointense to hypointense in relation to normal-appearing white matter on DWI, whereas most PCNSLs appeared hyperintense to white matter on DWI. Peripheral diffusion restriction (lower ADC value than that of the normal-appearing white matter) with increased ADC in the central portion was noted in four TDLs (three open ring–enhanced TDLs and one rim-enhanced TDL) and in three PCNSLs with ring enhancement. Visual and Quantitative Analyses of ADC Histograms By visual assessment, ADC histograms of both PCNSLs and TDLs were widely and asymmetrically distributed. The ADCmin (± SD) values for all PCNSLs, atypical PCNSLs, and TDLs were 563 ± 90, 537 ± 99, and 684 ±
AJR:204, April 2015
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
Histogram Analysis of ADC Maps 101 × 10−6 mm2/s, respectively. A significant difference was found between all of the PCNSLs and the TDLs (p = 0.000) as well as between the atypical PCNSLs and TDLs (p = 0.000). No differences in the ADCmean and ADCmode were found between PCNSLs and TDLs. The SD was higher in PCNSLs than in TDLs when all PCNSLs were analyzed (p = 0.013), but there was no significant difference between the subgroup of atypical PCNSLs and TDLs. Thirteen atypical PCNSLs (13/16) and twelve TDLs (12/18) showed positive skewness, whereas 18 typical PCNSLs (18/31) that had severe edema showed negative skewness. However, no statistical differences in skewness were found between all PCNSLs and TDLs as well as between a subgroup of atypical PCNSLs and TDLs. The kurtosis of all the PCNSLs and TDLs was leptokurtic without a statistically significant difference. The results of the cumulative histogram revealed that the ADC5% and ADC10% values showed significant differences (p = 0.003 and p = 0.048). In PCNSLs, all ADC5% and ADC10% values were lower than in TDLs. Lower ADC5% and ADC10% values were consistently found in the atypical PCNSLs compared with those in the TDLs (p = 0.001 and p = 0.009, respectively). However, no statistical differences were found in the other cumulative histogram parameters, including ADC25%, ADC50%, ADC75%, and ADC90%. The ADC histogram and cumulative histogram parameters of all PCNSLs, atypical PCNSLs, and TDLs are summarized in Table 2. A representative case of atypical PCNSL and the ADC histogram is shown in Figure 1. According to ROC curve analyses, the ADCmin value was shown to be the best indicator for differentiating all of the PCNSLs and TDLs (AUC, 0.812; 95% CI, 0.696– 0.898). The optimum cutoff value was 599 × 10 −6 mm2 /s (sensitivity, 68.1%; specificity, 77.8%). There were four TDLs showing lower ADCmin than 599× 10 −6 mm2 /s, three of which (75%) showed open-ring enhancement with low ADC observed in the periphery. The ADCmin was also useful for discriminating atypical PCNSLs from TDLs (AUC, 0.863; 95% CI, 0.701–0.956) with an optimum cutoff value of 556 × 10 −6 mm2 /s (sensitivity, 81.3%; specificity, 88.9%). The AUCs of ADC5% and ADC10% were lower than those of ADCmin. The diagnostic performance of ADCmin, ADC5%, and ADC10% is summarized in Table 3. Representative cases of atypical PCNSL and TDL mimicking each other are shown in Figures 2 and 3.
TABLE 2: Differences of Histogram Parameters in All Primary CNS Lymphomas (PCNSLs), Atypical PCNSLs, and Tumefactive Demyelinating Lymphomas (TDLs) All PCNSLs
Atypical PCNSLs
TDLs
pa
pb
ADCmean
1193 ± 212
1068 ± 239
1152 ± 197
0.477
0.270
ADCmode
1179 ± 367
1012 ± 309
1115 ± 246
0.422
0.286
ADCmin
563 ± 90
537 ± 99
684 ± 101
0.000
0.000
Skewness
0.09 ± 0.54
0.30 ± 0.48
0.34 ± 0.49
0.088
0.837
Kurtosis
2.69 ± 0.77
3.01 ± 0.75
2.80 ± 0.84
0.616
0.436
SD
26.8 ± 8.37
22.33 ± 10.57
20.45 ± 10.49
0.013
0.606
ADC5%
763 ± 108
723 ± 103
860 ± 124
0.003
0.001
ADC10%
833 ± 133
782 ± 126
907 ± 135
0.048
0.009
Parameter
Note—Except for p, data are reported as mean ± SD. Unit for ADC value is × 10 −6 mm2 /s. ADC = apparent diffusion coefficient, ADCmean = mean ADC value, ADCmode = ADC value at mode (peak height position of histogram), ADCmin = minimum value of ADC histogram, ADC 5% = 5th-percentile value of cumulative ADC histogram, ADC10% = 10th-percentile value of cumulative ADC histogram. ap value of compared results of all PCNSLs and TDLs using Student t test. bp value of compared results of a subgroup of atypical PCNSLs and TDLs.
TABLE 3: Diagnostic Performance of Histogram Parameters for D ifferentiating Primary CNS Lymphomas (PCNSLs) From Tumefactive D emyelinating Lymphomas (TDLs) AUC
Cutoff (× 10 −6 mm2 /s)
0.812 (0.696–0.898)
599
ADC5%
0.708 (0.582–0.814)
776
51.1 (36.1–65.9) 72.2 (46.5–90.3)
ADC10%
0.644 (0.516–0.759)
876
59.6 (44.3–73.6) 55.6 (30.8–78.5)
0.863 (0.701–0.956)
556
81.3 (54.4–96.0) 88.9 (65.3–98.6)
Parameter
Sensitivity (%)
Specificity (%)
Comparison of all PCNSLs and TDLs ADCmin
68.1 (52.9–80.9) 77.8 (52.4–93.6)
Comparison of atypical PCNSLs and TDLs ADCmin ADC5%
0.774 (0.599–0.899)
776
62.5 (35.4–84.8) 72.2 (46.5–90.3)
ADC10%
0.733 (0.554–0.869)
829
62.5 (35.4–84.8) 61.1 (35.7–82.7)
Note—Data in parentheses are 95% CI. AUC = largest area under the ROC curve, ADC = apparent diffusion coefficient, ADCmin = minimum value of the ADC histogram, ADC 5% = 5th-percentile value of cumulative ADC histogram, ADC10% = 10th-percentile value of cumulative ADC histogram.
Discussion The incidence of PCNSL is increasing, in part due to the increase of patients with HIV infection and the increase in the number of people with organ transplants who are undergoing immunosuppressive therapy [25, 26]. Although conventional MRI is an established tool for characterizing typical PCNSL, it is sometimes difficult to differentiate PCNSL from TDL, especially when there are atypical MRI manifestations. To our knowledge, there have been only a limited number of reports regarding DWI of atypical PCNSLs, and no literature reports investigating the usefulness of ADC histogram analysis for discriminating PCNSLs, especially atypical PCNSLs, from TDLs. Our results suggest that A DCmin
and the low-percentile values, including ADC5% and ADC10%, from cumulative ADC histograms based on the T2-weighted imaging abnormalities, could help to differentiate PCNSLs from TDLs and could also be useful for discriminating atypical PCNSLs from TDLs. However, the ADCmean, ADCmode, and other parameters of histogram distribution, particularly kurtosis and skewness, have limited diagnostic value. DWI reflects the free diffusion of water molecules in biologic tissue [27]. Because PCNSLs are highly cellular tumors with a high nuclear-to-cytoplasmic ratio, diffusion restriction has been reported as one of the important MRI findings [28]. Previous studies suggest that PCNSLs reveal more
AJR:204, April 2015 831
restricted diffusion and lower ADC values compared with high-grade gliomas [18, 19, 28]. In the study by Calli et al. [19], the ADCmin of PCNSL was reported as 510 ± 90 × 10 −6 mm2 /s, which helped to distinguish PCNSLs from other intracerebral malignancies. In our study, ADCmin and the low-percentile values determined from the cumulative ADC histogram (ADC5% and ADC10%) were found to be low in PCNSL, in which the ADCmin was 563 ± 90 ×10 −6 mm2 /s, which was in accordance with the results of previous literature reports. The typical DWI feature of TDLs is their variable ADC values [11, 14, 29]. It has been reported that most TDLs have elevated ADC values because of their vasogenic edema as well myelin destruction with axonal preservation [30]. Similar findings were also noted in our study. However, acute demyelinating lesions may show areas of reduced ADC values that are thought to be caused by intramyelinic edema or myelin vacuolation [31]. Contrast enhancement is considered a hallmark of biologically active lesions and reflects a breakdown in the blood-brain barrier [29]. In our TDL group, mostly very low ADCmin was observed in the enhancing portion of lesions in patients with the openring enhancement pattern, and peripheral restricted diffusion was the primary manifestation seen on DWI. Nevertheless, we found that the ADCmin of TDL was not as low as that of PCNSL and was therefore useful for discriminating PCNSLs because of its good diagnostic performance. Although the diagnostic performance of ADC5% and ADC10% does not seem to be as successful as that of ADCmin, the cumulative percentiles derived from a cumulative histogram may be more reliable parameters than the minimum values because they may be less affected by misregistration artifacts or imperfect delineation of lesion boundaries [20]. MRI features of atypical PCNSLs are more heterogeneous compared with homogeneously enhancing typical PCNSLs. In nine of the 16 atypical PCNSLs in our patient groups, it was not easy to delineate the tumor portions because they were obscured by edema. Therefore, in our study, we included all of the high-signal-intensity abnormalities seen on T2-weighted imaging for the ADC histogram analysis, including necrosis and edema, to obtain a more accurate evaluation of heterogeneous lesions. All PCNSLs showed higher SD than TDLs. We think that more severe ede-
832
A
B
C
D
160 140 120 Frequency
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
Lu et al.
100 80 60 40 20 0 600
800 ADCmin
1000 1200 1400 1600 1800 ADC5% ADC10% ADC Value (10−6 mm2/s)
2000
2200
2400
E Fig. 3—37-year-old woman with biopsy-proven tumefactive demyelinating lesion mimicking primary CNS lymphoma (PCNSL). A, T2-weighted image shows well-defined lesion in right frontal lobe with slightly high signal intensity and relatively little surrounding vasogenic edema. B, Contrast-enhanced T1-weighted image shows obvious patchy enhancement. C and D, Diffusion-weighted image (C) and corresponding apparent diffusion coefficient (ADC) map (D) reveal slightly decreased diffusion within lesion. E, ADC histogram based on entire lesion as well as edema shows wide distribution. Minimum ADC (ADCmin), 5th-percentile ADC (ADC 5%), and 10th-percentile ADC (ADC10%) values were 771, 1092, and 1198 × 10 −6 mm2 /s, respectively, all of which are higher than those of PCNSL.
AJR:204, April 2015
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
Histogram Analysis of ADC Maps ma, hemorrhage, and necrosis contribute to the greater variation of ADC values in PCNSLs. Other distribution parameters, including skewness and kurtosis, did not show statistically significant differences. Most TDLs (12/18) and atypical PCNSLs (13/16) showed positive skewness. Eighteen of 31 typical PCNSLs showed negative skewness, which may be due to the more severe edema in these cases. The low ADC values of the histogram may correspond to the densely packed lymphoid cells in PCNSLs or to the acute demyelinating plaque in TDLs, which usually appear as the enhancing portion of the lesions. The high ADC values are possibly associated with a mixture of viable and necrotic cells and edema. The sensitivity and specificity for ADCmin to differentiate all PCNSLs from TDLs were low (sensitivity, 68.1%; specificity, 77.8%). However, those to differentiate atypical PCNSLs from TDLs were high (sensitivity, 81.3%; specificity, 88.9%). It is unclear why there is a discrepancy between the two comparisons. We speculate that perfusion effect may have influenced the ADC value. Increased perfusion has been reported to cause increased ADC in the liver and also in brain tumors [25, 32]. Although perfusion in PCNSLs is not high, they still have increased perfusion. Solid-enhancing masses in typical PCNSLs may have higher perfusion than that of atypical PCNSLs, which usually have heterogeneous enhancement, and this may explain the higher ADCmin in the entire PCNSLs group than in the atypical PCNSLs group. Our findings suggest that histogram analysis of ADC is especially useful for differentiating atypical PCNSLs from TDLs, which is sometimes challenging in clinical practice. Nonenhancing PCNSL usually causes a diagnostic dilemma. In our previous study, we found that 1H-MR spectroscopy of nonenhancing PCNSLs was nonspecific and could not help in discriminating them from TDLs [26]. However, the ADCmin, ADC5%, and ADC10% of the nonenhancing PCNSLs in our series were lower than those of the TDLs, which indicates that ADC histogram analysis might be helpful in making this differentiation. However, because there was only one case in our study, further studies will be necessary to substantiate our findings. We found that ADC histograms were asymmetric, had broad distribution, and were bimodal or even multimodal, which agreed with previous reports [33]. Therefore,
we believe that the gaussian mixture modeling used in our study could more accurately provide better curve fitting and reflect the heterogeneity of the lesions in their entirety. There are several limitations to our study. First, the number of patients was relatively small and it was a retrospective study. More PCNSLs, especially atypical PCNSLs, and TDLs will be required in future studies to strengthen the statistical power. Second, accurately defining the boundary of lesions is challenging, especially when diffuse confluent lesions are shown. However, given the large number of pixel data included in the histogram of an entire lesion, the imperfect delineation of lesion boundaries is relatively unimportant according to the results of previous studies. Last, coregistration of T2weighted imaging with the ADC map was required in our study. Although motion correction had been performed and ROIs were carefully drawn, there is still the possibility of misregistration artifacts and the inclusion of erroneous ADC values at the lesion boundary. In that respect, cumulative histogram parameters, such as ADC5% and ADC10%, may be more promising. In conclusion, our results suggest that ADC histogram analysis based on an entire lesion could help to discriminate PCNSLs and TDLs and may be particularly useful for differentiating the subgroup of atypical PCNSLs from TDLs. The ADCmin and the low-percentile values of the cumulative ADC histogram, including ADC5% and ADC10%, may be potentially promising parameters. References 1. Küker W, Nägele T, Korfel A, et al. Primary central nervous system lymphomas (PCNSL): MRI features at presentation in 100 patients. J Neurooncol 2005; 72:169–177 2. Zacharia TT, Law M, Naidich TP, Leeds NE. Central nervous system lymphoma characterization by diffusion-weighted imaging and MR spectroscopy. J Neuroimaging 2008; 18:411–417 3. Haldorsen IS, Espeland A, Larsson EM. Central nervous system lymphoma: characteristic findings on traditional and advanced imaging. AJNR 2011; 32:984–992 4. Haldorsen IS, Kråkenes J, Krossnes BK, Mella O, Espeland A. CT and MR imaging features of primary central nervous system lymphoma in Norway, 1989–2003. AJNR 2009; 30:744–751 5. Tang YZ, Booth TC, Bhogal P, Malhotra A, Wilhelm T. Imaging of primary central nervous system lymphoma. Clin Radiol 2011; 66:768–777 6. Erdag N, Bhorade RM, Alberico RA, Yousuf N,
Patel MR. Primary lymphoma of the central nervous system: typical and atypical CT and MR imaging appearances. AJR 2001; 176:1319–1326 7. Jahnke K, Schilling A, Heidenreich J, et al. Radiologic morphology of low-grade primary central nervous system lymphoma in immunocompetent patients. AJNR 2005; 26:2446–2454 8. Lachenmayer ML, Blasius E, Niehusmann P, et al. Non-enhancing primary CNS lymphoma. J Neurooncol 2011; 101:343–344 9. Trendelenburg G, Zimmer C, Forschler A, Stadelmann C, Zschenderlein R. Atypical appearance of a primary central nervous system lymphoma. Arch Neurol 2006; 63:908–909 10. Ragel BT, Fassett DR, Baringer JR, Browd SR, Dailey AT. Decompressive hemicraniectomy for tumefactive demyelination with transtentorial herniation: observation. Surg Neurol 2006; 65:582–583 11. Given CA 2nd, Stevens BS, Lee C. The MRI appearance of tumefactive demyelinating lesions. AJR 2004; 182:195–199 12. Saindane AM, Cha S, Law M, Xue X, Knopp EA, Zagzag D. Proton MR spectroscopy of tumefactive demyelinating lesions. AJNR 2002; 23:1378–1386 13. Cha S, Pierce S, Knopp EA, et al. Dynamic contrast-enhanced T2*-weighted MR imaging of tumefactive demyelinating lesions. AJNR 2001; 22:1109–1116 14. Al-Okaili RN, Krejza J, Wang S, Woo JH, Melhem ER. Advanced MR imaging techniques in the diagnosis of intraaxial brain tumors in adults. RadioGraphics 2006; 26(suppl 1):S173–S189 15. Schroeder PC, Post MJ, Oschatz E, Stadler A, Bruce-Gregorios J, Thurnher MM. Analysis of the utility of diffusion-weighted MRI and apparent diffusion coefficient values in distinguishing central nervous system toxoplasmosis from lymphoma. Neuroradiology 2006; 48:715–720 16. Horger M, Fenchel M, Nägele T, et al. Water diffusivity: comparison of primary CNS lymphoma and astrocytic tumor infiltrating the corpus callosum. AJR 2009; 193:1384–1387 17. Camacho DL, Smith JK, Castillo M. Differentiation of toxoplasmosis and lymphoma in AIDS patients by using apparent diffusion coefficients. AJNR 2003; 24:633–637 18. Doskaliyev A, Yamasaki F, Ohtaki M, et al. Lymphomas and glioblastomas: differences in the apparent diffusion coefficient evaluated with high b-value diffusion-weighted magnetic resonance imaging at 3T. Eur J Radiol 2012; 81:339–344 19. Calli C, Kitis O, Yunten N, Yurtseven T, Islekel S, Akalin T. Perfusion and diffusion MR imaging in enhancing malignant cerebral tumors. Eur J Radiol 2006; 58:394–403 20. Kang Y, Choi SH, Kim YJ, et al. Gliomas: histogram analysis of apparent diffusion coefficient
AJR:204, April 2015 833
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
Lu et al. maps with standard- or high-b-value diffusionweighted MR imaging—correlation with tumor grade. Radiology 2011; 261:882–890 21. Nowosielski M, Recheis W, Goebel G, et al. ADC histograms predict response to anti-angiogenic therapy in patients with recurrent high-grade glioma. Neuroradiology 2011; 53:291–302 22. Al-Okaili RN, Krejza J, Woo JH, et al. Intraaxial brain masses: MR imaging-based diagnostic strategy—initial experience. Radiology 2007; 243:539–550 23. Toh CH, Wei KC, Ng SH, Wan YL, Castillo M, Lin CP. Differentiation of tumefactive demyelinating lesions from high-grade gliomas with the use of diffusion tensor imaging. AJNR 2012; 33:846–851 24. Yap KK, Sutherland T, Liew E, Tartaglia CJ, Pang M, Trost N. Magnetic resonance features of pri-
mary central nervous system lymphoma in the immunocompetent patient: a pictorial essay. J Med Imaging Radiat Oncol 2012; 56:179–186 25. Cohen AD, LaViolette PS, Prah M, et al. Effects of perfusion on diffusion changes in human brain tumors. J Magn Reson Imaging 2013; 38:868–875 26. Lu SS, Kim SJ, Kim HS, et al. Utility of proton MR spectroscopy for differentiating typical and atypical primary central nervous system lymphomas from tumefactive demyelinating lesions. AJNR 2014; 35:270–277 27. Koh DM, Collins DJ. Diffusion-weighted MRI in the body: applications and challenges in oncology. AJR 2007; 188:1622–1635 28. Guo AC, Cummings TJ, Dash RC, Provenzale JM. Lymphomas and high-grade astrocytomas: comparison of water diffusibility and histologic characteristics. Radiology 2002; 224:177–183
29. Balashov KE, Lindzen E. Acute demyelinating lesions with restricted diffusion in multiple sclerosis. Mult Scler 2012; 18:1745–1753 30. Tievsky AL, Ptak T, Farkas J. Investigation of apparent diffusion coefficient and diffusion tensor anisotropy in acute and chronic multiple sclerosis lesions. AJNR 1999; 20:1491–1499 31. Abou Zeid N, Pirko I, Erickson B, et al. Diffusionweighted imaging characteristics of biopsy-proven demyelinating brain lesions. Neurology 2012; 78:1655–1662 32. Hollingsworth KG, Lomas DJ. Influence of perfusion on hepatic MR diffusion measurement. NMR Biomed 2006; 19:231–235 33. Pope WB, Lai A, Mehta R, et al. Apparent diffusion coefficient histogram analysis stratifies progressionfree survival in newly diagnosed bevacizumab- treated glioblastoma. AJNR 2011; 32:882–889
F O R YO U R I N F O R M AT I O N
Unique customized medical search engine service from ARRS! ARRS GoldMiner® is a keyword- and concept-driven search engine that provides instant access to radiologic images published in peer-reviewed journals. ARRS members earn 0.5 CME credits for each self-directed search. For more information, visit goldminer.arrs.org.
834
AJR:204, April 2015
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
Neuroradiology/Head and Neck Imaging Original Research
Histogram Analysis of Apparent Diffusion Coefficient Maps for Differentiating Primary CNS Lymphomas From Tumefactive Demyelinating Lesions Shan Shan Lu1,2 Sang Joon Kim1 Namkug Kim1 Ho Sung Kim1 Choong Gon Choi1 Young Min Lim 3 Lu SS, Kim SJ, Kim N, Kim HS, Choi CG, Lim YM
Keywords: apparent diffusion coefficient (ADC), demyelinating disease, diffusion MRI, histogram analysis, primary CNS lymphoma (PCNSL) DOI:10.2214/AJR.14.12677 Received February 10, 2014; accepted after revision June 11, 2014. 1 Department of Radiology and Research Institute of Radiology, Asan Medical Center, University of Ulsan College of Medicine, 88 Olympic-ro 43-gil, Songpa-gu, Seoul 138-736, Korea. Address correspondence to S. J. Kim ([email protected]). 2
Department of Radiology, The First Affiliated Hospital of Nanjing Medical University, Nanjing, Jiangsu Province, China.
3
Department of Neurology, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Korea.
AJR 2015; 204:827–834 0361–803X/15/2044–827 © American Roentgen Ray Society
OBJECTIVE. This study intended to investigate the usefulness of histogram analysis of apparent diffusion coefficient (ADC) maps for discriminating primary CNS lymphomas (PCNSLs), especially atypical PCNSLs, from tumefactive demyelinating lesions (TDLs). MATERIALS AND METHODS. Forty-seven patients with PCNSLs and 18 with TDLs were enrolled in our study. Hyperintense lesions seen on T2-weighted images were defined as ROIs after ADC maps were registered to the corresponding T2-weighted image. ADC histograms were calculated from the ROIs containing the entire lesion on every section and on a voxel-by-voxel basis. The ADC histogram parameters were compared among all PCNSLs and TDLs as well as between the subgroup of atypical PCNSLs and TDLs. ROC curves were constructed to evaluate the diagnostic performance of the histogram parameters and to determine the optimum thresholds. RESULTS. The differences between the PCNSLs and TDLs were found in the minimum ADC values (ADCmin) and in the 5th and 10th percentiles (ADC5% and ADC10%) of the cumulative ADC histograms. However, no statistical significance was found in the mean ADC value or in the ADC value concerning the mode, kurtosis, and skewness. The ADCmin, ADC5%, and ADC10% were also lower in atypical PCNSLs than in TDLs. ADCmin was the best indicator for discriminating atypical PCNSLs from TDLs, with a threshold of 556 × 10 −6 mm2 /s (sensitivity, 81.3 %; specificity, 88.9%). CONCLUSION. Histogram analysis of ADC maps may help to discriminate PCNSLs from TDLs and may be particularly useful in differentiating atypical PCNSLs from TDLs.
N
ow frequently encountered in both immunocompetent and immunocompromised patients, primary CNS lymphomas (PCNSLs) constitute from 1% to 6% of primary brain neoplasms [1, 2]. In patients with normal immunity, PCNSL usually presents as a well-demarcated solitary lesion without necrosis or hemorrhage and with marked homogeneous enhancement, both of which are considered to be typical imaging features [3– 5]. However, many atypical radiologic presentations have been reported, such as necrosis, hemorrhage, variably heterogeneous enhancement, and sometimes even no enhancement or as a diffuse lesion without mass formation [6–9]. Correct diagnosis of typical PCNSLs is usually not difficult in routine clinical practice. However, when unusual MRI manifestations are present, differentiating atypical PCNSLs from tumefactive demyelinating lesions (TDLs) can sometimes be challenging,
both clinically and radiologically. The incidence of TDL is estimated as 0.3 per 100,000 per year [10]. They can cause symptoms mimicking brain tumors and commonly have heterogeneous enhancement [11]. However, accurate recognition of PCNSL according to the imaging criteria is important to avoid inappropriate use of steroid medication and to ensure timely and adequate treatment. Many advanced imaging techniques have been used for distinguishing between PCNSLs and TDLs. Choline/creatine (Cho/ Cr) ratios and choline/N-acetyl aspartate (Cho/NAA) ratios derived from MR spectroscopy may differ between the two diseases, although there are still some overlaps [2, 12]. Similarly, in dynamic susceptibility contrast T2*-weighted perfusion MRI, PCNSLs tend to have higher relative cerebral blood volume than TDLs, although sometimes there is only a slight difference [13]. The role of diffusion-weighted imaging (DWI) with apparent diffusion coeffi-
AJR:204, April 2015 827
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
Lu et al. cient (ADC) maps is currently being actively investigated. PCNSL has been reported to have a tendency toward low ADC because of its high cellularity, which helps to differentiate it from toxoplasmosis or gliomas [14– 16]. However, to our knowledge, the role of DWI in distinguishing PCNSLs from TDLs has not yet been reported, and there are only a limited number of reports evaluating ADC in the subgroup of atypical PCNSLs. To measure ADC in brain lesions, most of the previous studies used ROI analysis, which was based only on the representative part of the tumor [17–19]. However, selecting a location is subjective and may be difficult when there are atypical imaging features. Recently, ADC histogram analysis has been found to be useful for grading gliomas and predicting the treatment response and progression-free survival in patients with glioblastomas [20, 21]. Given that histogram analysis includes an entire lesion, it may reveal the heterogeneity of lesions and thus ensure more accurate discrimination. In our study, we intended to evaluate the clinical utility of histogram analysis of ADC maps for differentiating PCNSLs, especially atypical PCNSLs, from TDLs, and we further investigated the diagnostic performance of ADC histograms for this discrimination.
regression of the lesions and clinical symptoms during follow-up after corticosteroid treatment. The final diagnosis of TDL was made on the basis of either the histopathology findings or a strong clinical suspicion. Among the TDLs, five patients were confirmed by stereotactic biopsy and 13 were diagnosed by clinical follow-up and repeat MRI. PCNSL patients were further divided into two subgroups on the basis of conventional MRI: typical PCNSLs had a nodular pattern with intense homogeneous enhancement and a well-defined margin accompanied by surrounding edema (n = 31), and atypical PCNSLs showed necrosis, hemorrhage, or atypical enhancing patterns on contrast-enhanced T1-weighted imaging, including patchy infiltrative enhancement, streaky infiltrative enhancement, ring enhancement, or even lack of enhancement (n = 16) [24].
Image Acquisition All of the images were obtained on a 3-T MR system (Achieva, Philips Healthcare) with an eight-channel head coil. The conventional protocols included axial T1-weighted (TR/TE, 500/10; slice thickness, 5 mm; slice gap, 2 mm; FOV, 230 × 183 mm; matrix, 256 × 183; number of signals acquired, 1; and voxel resolution, 0.90 × 1.00 × 5.00 mm), T2-weighted fast spinecho (TR/TE, 3000/80; slice thickness, 5 mm; slice gap, 2 mm; FOV, 230 × 184 mm; matrix, 400 × 255; number of signals acquired, 1; and voxel resolution, 0.58 × 0.72 × 5.00 mm), FLAIR (TR/TE, 10,000/125; slice thickness, 5 mm; slice gap, 2 mm; inversion time, 2200 ms; FOV, 230 × 182 mm; matrix, 352 × 230; number of signals acquired, 1; and voxel resolution, 0.65 × 0.90 × 5.00 mm), and contrast-enhanced T1-weighted 3D fast
Materials and Methods Patients This retrospective study was approved by our institutional review board, and the requirement for informed consent was waived. We reviewed our institutional database for the time period between April 2007 and May 2012 and selected 47 PCNSL patients (33 men and 14 women; age, 29–82 years) and 18 TDL patients (10 men and eight women; age, 22–66 years). Our inclusion criteria were histopathologic diagnosis of PCNSLs by surgical resection or stereotactic biopsy, histopathologically proven or clinically diagnosed TDLs, and raw DWI data available with good quality along with the conventional MRI results. Eighteen PCNSLs and five TDLs were excluded because of inadequate DWI, including incomplete raw data and data with motion artifact or image distortion. The clinical diagnosis of TDLs was based on the following criteria [12, 13, 22, 23]: acute or subacute onset of neurologic symptoms and signs; at least one brain lesion with the longest diameter more than 2 cm on MRI; supportive CSF analysis, including oligoclonal bands, IgG index, myelin basic protein, or supportive somatosensory-evoked potential tests; no evidence of systemic illness, vasculitis, toxic or metabolic disease, or CNS infection; and documented
828
A
B
C
D
Fig. 1—64-year-old man with atypical primary CNS lymphoma. A and B, Multifocal hyperintensity lesions are seen on FLAIR images involving bilateral frontal cortex and subcortical white matter (A) and with patchy infiltrative enhancement in left frontal lobe (B). C and D, Diffusion-weighted image (C) and corresponding apparent diffusion coefficient (ADC) map (D) show obvious diffusion restriction (arrow, D). (Fig. 1 continues on next page)
AJR:204, April 2015
Histogram Analysis of ADC Maps
250 Frequency
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
300
200 150 100 50 0 400 ADCmin
600 800 ADC5% ADC10%
1000
1200
1400
1600
1800
ADC Value (10 mm /s) −6
2
E Fig. 1 (continued)—64-year-old man with atypical primary CNS lymphoma. E, ADC histogram of lesions is asymmetric, with low minimum ADC (ADCmin), 5th-percentile ADC (ADC 5%), and 10th-percentile ADC (ADC10%) of 472, 622, and 648 × 10 −6 mm2 /s, respectively (arrows).
field echo (TR/TE, 9.8/4.6; slice thickness, 1 mm; FOV, 224 cm; matrix, 224 × 224; number of signals acquired, 1; and voxel resolution, 1.0 × 1.0 × 1.0 mm) imaging with transverse, coronal, and sagittal planes after IV injection of 0.1 mmol/kg of gadoterate meglumine (Dotarem, Guerbet). Echo-planar DWI was performed in the axial plane with b = 0 and 1000 s/mm2. Other parameters included TR/TE, 3000/56; slice thickness, 5 mm; slice gap, 2 mm; FOV, 250 × 250 mm; matrix, 128 ×125; number of signals acquired, 1; and voxel resolution, 1.95 × 2.00 × 5.00 mm. DWI was performed in three orthogonal directions of diffusion sensitizing gradients and combined into isotropic images.
er in-plane resolution than that of DWI (1.95 × 2 mm), and therefore ADC maps were coregistered to T2-weighted imaging. Motion correction was performed to correct a rigid body motion to the time series data. This was done automatically in the Nordic ICE software. The high-signal-intensity area on T2-weighted imaging was identified as the outermost lesion margin. ROIs were manually drawn around the entire lesions on all of the axial sections from the coregistered T2-weighted–ADC images with reference to the underlying T2-weighted image. Necrosis, edema, and hemorrhage were all included and the contrast-enhanced portion seen on contrast-enhanced T1-weighted imaging was always included. The ADC values were calculated on a
voxel-by-voxel basis from all of the ROIs and recorded in a separated file using NordicICE software. The files were then transported to an inhouse software developed using Matlab 2010b (MathWorks) in a blinded manner. The histogram of the ADC values was plotted on the x-axis with a predefined total bin number of 500. The y-axis was recorded as the frequency of each bin. To fit the ADC histograms, which are usually distributed broadly and asymmetrically in tumors, we used gaussian mixture modeling in this study. The gaussian mixture number was empirically set as 2 or 3 for generating a suitable histogram fitting curve. All of the gaussian mixture number and fitting curves were checked and verified by a neuroradiologist with 20 years of experience to ensure the best curve fitting. We evaluated the various representative parameters, including the mean ADC value (A DCmean); mode ADC value (ADCmode, the peak height position of the histogram); minimum ADC value (ADCmin); and skewness, kurtosis, and SD of the histograms. We also measured the six cumulative histogram parameters including the 5th (ADC5%), 10th (ADC10%), 25th (ADC25%), 50th (ADC50%), 75th (ADC75%), and 90th (ADC90%) percentiles. The nth percentile was the point at which n% of the voxel values that form the histogram were found to the left [20].
Statistical Analysis All of the data were reported as the mean ± SD. Conventional MRI features of PCNSLs and TDLs were compared using the Pearson chi-square test.
Image Processing and Analysis Conventional MRI features, including the number of lesions, enhancing pattern, hemorrhage, and necrosis, were analyzed. The enhancing patterns were classified into six types: nodular pattern, lesions with well-defined margins and homogeneous enhancement; ring pattern, complete rimlike enhancement in the periphery without enhancement in the center; open-ring pattern, incomplete rimlike enhancement in the periphery; patchy infiltrative pattern, ill-defined or irregularly marginated lesions with heterogeneous enhancement; streaky infiltrative pattern, linear or dotlike enhancement; and nonenhancing pattern, abnormal signal intensity on T2-weighted imaging without definite contrast enhancement. All of the raw DWI data were transferred from the MRI scanner to an independent PC for generating ADC maps and image coregistration using a dedicated software package (NordicICE, NordicNeuroLab). ADC maps were generated using the monoexponential expression for diffusion attenuation after spatial and temporal smoothing. T2-weighted imaging (0.58 × 0.72 mm) had high-
A
B
Fig. 2—29-year-old man with atypical primary CNS lymphoma (PCNSL) mimicking tumefactive demyelinating lesion (TDL). A and B, Confluent hyperintensity lesions are observed on FLAIR image (A) in both frontal periventricular white matter and parietooccipital white matter and with streaky infiltrative enhancement seen on contrast-enhanced T1-weighted image (B). (Fig. 2 continues on next page)
AJR:204, April 2015 829
The ADC histogram parameters were compared using the Student t test in all of the PCNSLs and TDLs and then between atypical PCNSLs and TDLs. ROC curve analyses were performed to evaluate the diagnostic performance of each histogram parameter and to determine the optimum thresholds for discriminating PCNSLs, especially atypical PCNSLs, from TDLs. The AUC, sensitivity, and specificity were further calculated using the ROC curve. All analyses were performed using commercially available statistical software packages (SPSS, version 13.0, IBM, or MedCalc, version 12.3.0, MedCalc Software). The p value was two-sided and a value of less than 0.05 was considered statistically significant.
C
D
350 300 250 Frequency
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
Lu et al.
200 150 100 50 0 500 600 700 800 ADCmin ADC5% ADC10%
900
1000
1100
1200
1300
1400
1500
ADC Value (10 mm /s) −6
2
E Fig. 2 (continued)—29-year-old man with atypical primary CNS lymphoma (PCNSL) mimicking tumefactive demyelinating lesion (TDL). C and D, Diffusion-weighted images show prominent increased signal (C) with partially decreased apparent diffusion coefficient (ADC) in peripheral portion (arrows, D). E, ADC histogram shows low minimum ADC (ADCmin), 5th-percentile ADC (ADC 5%), and 10th-percentile ADC (ADC10%) of 556, 669, and 697 × 10 −6 mm2 /s, respectively, suggesting PCNSL rather than TDL.
TABLE 1: Conventional MRI Features of Primary CNS Lymphomas (PCNSLs) and Tumefactive Demyelinating Lymphomas (TDLs) Feature
PCNSLs (n = 47)
TDLs (n = 18)
p
Single lesion
26 (55.3)
5 (27.8)
0.047
Multiple focal lesions
15 (31.9)
9 (50.0)
0.176
Diffuse lesions
6 (12.8)
4 (22.2)
0.575
31 (66.0)
0 (0.0)
0.000
No. of lesions
Enhancing pattern Nodular Ring
4 (8.5)
1 (5.6)
1.000
Open ring
0 (0.0)
6 (33.3)
0.000
Patchy infiltrative
7 (14.9)
3 (22.2)
1.000
Streaky infiltrative
4 (8.5)
1 (0.0)
1.000
1 (2.1)
7 (38.9)
0.000
Necrosis
No enhancement
5 (10.6)
6 (33.3)
0.070
Hemorrhage
4 (8.5)
1 (5.6)
1.000
Note—Data in parentheses are percentages.
830
Results Visual Assessment of Conventional MRI and DWI On conventional MRI, a single lesion was observed in 26 PCNSL patients and five TDL patients, and multiple lesions were found in 15 PCNSL patients and in nine TDL patients. Diffusely distributed lesions were observed in the remaining six PCNSL patients and four TDL patients. In nine of 16 (56.3%) atypical PCNSL patients, lesions could not be discriminated from edema. The following atypical features were found in a substantial proportion of the atypical PCNSL patients: necrosis (n = 5), hemorrhage (n = 4), ring enhancement (n = 4), patchy infiltrative enhancement (n = 7), streaky infiltrative enhancement (n = 4), or absent contrast enhancement (n = 1). Eleven of 18 TDLs showed varying enhancement, including an open-ring pattern (n = 6), rimlike pattern (n = 1), patchy infiltrative pattern (n = 3), and streaky infiltrative pattern (n = 1). No obvious enhancement was observed in the other seven TDLs. The conventional MRI manifestations of PCNSLs and TDLs are summarized and compared in Table 1. Most TDLs appeared predominantly isointense to hypointense in relation to normal-appearing white matter on DWI, whereas most PCNSLs appeared hyperintense to white matter on DWI. Peripheral diffusion restriction (lower ADC value than that of the normal-appearing white matter) with increased ADC in the central portion was noted in four TDLs (three open ring–enhanced TDLs and one rim-enhanced TDL) and in three PCNSLs with ring enhancement. Visual and Quantitative Analyses of ADC Histograms By visual assessment, ADC histograms of both PCNSLs and TDLs were widely and asymmetrically distributed. The ADCmin (± SD) values for all PCNSLs, atypical PCNSLs, and TDLs were 563 ± 90, 537 ± 99, and 684 ±
AJR:204, April 2015
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
Histogram Analysis of ADC Maps 101 × 10−6 mm2/s, respectively. A significant difference was found between all of the PCNSLs and the TDLs (p = 0.000) as well as between the atypical PCNSLs and TDLs (p = 0.000). No differences in the ADCmean and ADCmode were found between PCNSLs and TDLs. The SD was higher in PCNSLs than in TDLs when all PCNSLs were analyzed (p = 0.013), but there was no significant difference between the subgroup of atypical PCNSLs and TDLs. Thirteen atypical PCNSLs (13/16) and twelve TDLs (12/18) showed positive skewness, whereas 18 typical PCNSLs (18/31) that had severe edema showed negative skewness. However, no statistical differences in skewness were found between all PCNSLs and TDLs as well as between a subgroup of atypical PCNSLs and TDLs. The kurtosis of all the PCNSLs and TDLs was leptokurtic without a statistically significant difference. The results of the cumulative histogram revealed that the ADC5% and ADC10% values showed significant differences (p = 0.003 and p = 0.048). In PCNSLs, all ADC5% and ADC10% values were lower than in TDLs. Lower ADC5% and ADC10% values were consistently found in the atypical PCNSLs compared with those in the TDLs (p = 0.001 and p = 0.009, respectively). However, no statistical differences were found in the other cumulative histogram parameters, including ADC25%, ADC50%, ADC75%, and ADC90%. The ADC histogram and cumulative histogram parameters of all PCNSLs, atypical PCNSLs, and TDLs are summarized in Table 2. A representative case of atypical PCNSL and the ADC histogram is shown in Figure 1. According to ROC curve analyses, the ADCmin value was shown to be the best indicator for differentiating all of the PCNSLs and TDLs (AUC, 0.812; 95% CI, 0.696– 0.898). The optimum cutoff value was 599 × 10 −6 mm2 /s (sensitivity, 68.1%; specificity, 77.8%). There were four TDLs showing lower ADCmin than 599× 10 −6 mm2 /s, three of which (75%) showed open-ring enhancement with low ADC observed in the periphery. The ADCmin was also useful for discriminating atypical PCNSLs from TDLs (AUC, 0.863; 95% CI, 0.701–0.956) with an optimum cutoff value of 556 × 10 −6 mm2 /s (sensitivity, 81.3%; specificity, 88.9%). The AUCs of ADC5% and ADC10% were lower than those of ADCmin. The diagnostic performance of ADCmin, ADC5%, and ADC10% is summarized in Table 3. Representative cases of atypical PCNSL and TDL mimicking each other are shown in Figures 2 and 3.
TABLE 2: Differences of Histogram Parameters in All Primary CNS Lymphomas (PCNSLs), Atypical PCNSLs, and Tumefactive Demyelinating Lymphomas (TDLs) All PCNSLs
Atypical PCNSLs
TDLs
pa
pb
ADCmean
1193 ± 212
1068 ± 239
1152 ± 197
0.477
0.270
ADCmode
1179 ± 367
1012 ± 309
1115 ± 246
0.422
0.286
ADCmin
563 ± 90
537 ± 99
684 ± 101
0.000
0.000
Skewness
0.09 ± 0.54
0.30 ± 0.48
0.34 ± 0.49
0.088
0.837
Kurtosis
2.69 ± 0.77
3.01 ± 0.75
2.80 ± 0.84
0.616
0.436
SD
26.8 ± 8.37
22.33 ± 10.57
20.45 ± 10.49
0.013
0.606
ADC5%
763 ± 108
723 ± 103
860 ± 124
0.003
0.001
ADC10%
833 ± 133
782 ± 126
907 ± 135
0.048
0.009
Parameter
Note—Except for p, data are reported as mean ± SD. Unit for ADC value is × 10 −6 mm2 /s. ADC = apparent diffusion coefficient, ADCmean = mean ADC value, ADCmode = ADC value at mode (peak height position of histogram), ADCmin = minimum value of ADC histogram, ADC 5% = 5th-percentile value of cumulative ADC histogram, ADC10% = 10th-percentile value of cumulative ADC histogram. ap value of compared results of all PCNSLs and TDLs using Student t test. bp value of compared results of a subgroup of atypical PCNSLs and TDLs.
TABLE 3: Diagnostic Performance of Histogram Parameters for D ifferentiating Primary CNS Lymphomas (PCNSLs) From Tumefactive D emyelinating Lymphomas (TDLs) AUC
Cutoff (× 10 −6 mm2 /s)
0.812 (0.696–0.898)
599
ADC5%
0.708 (0.582–0.814)
776
51.1 (36.1–65.9) 72.2 (46.5–90.3)
ADC10%
0.644 (0.516–0.759)
876
59.6 (44.3–73.6) 55.6 (30.8–78.5)
0.863 (0.701–0.956)
556
81.3 (54.4–96.0) 88.9 (65.3–98.6)
Parameter
Sensitivity (%)
Specificity (%)
Comparison of all PCNSLs and TDLs ADCmin
68.1 (52.9–80.9) 77.8 (52.4–93.6)
Comparison of atypical PCNSLs and TDLs ADCmin ADC5%
0.774 (0.599–0.899)
776
62.5 (35.4–84.8) 72.2 (46.5–90.3)
ADC10%
0.733 (0.554–0.869)
829
62.5 (35.4–84.8) 61.1 (35.7–82.7)
Note—Data in parentheses are 95% CI. AUC = largest area under the ROC curve, ADC = apparent diffusion coefficient, ADCmin = minimum value of the ADC histogram, ADC 5% = 5th-percentile value of cumulative ADC histogram, ADC10% = 10th-percentile value of cumulative ADC histogram.
Discussion The incidence of PCNSL is increasing, in part due to the increase of patients with HIV infection and the increase in the number of people with organ transplants who are undergoing immunosuppressive therapy [25, 26]. Although conventional MRI is an established tool for characterizing typical PCNSL, it is sometimes difficult to differentiate PCNSL from TDL, especially when there are atypical MRI manifestations. To our knowledge, there have been only a limited number of reports regarding DWI of atypical PCNSLs, and no literature reports investigating the usefulness of ADC histogram analysis for discriminating PCNSLs, especially atypical PCNSLs, from TDLs. Our results suggest that A DCmin
and the low-percentile values, including ADC5% and ADC10%, from cumulative ADC histograms based on the T2-weighted imaging abnormalities, could help to differentiate PCNSLs from TDLs and could also be useful for discriminating atypical PCNSLs from TDLs. However, the ADCmean, ADCmode, and other parameters of histogram distribution, particularly kurtosis and skewness, have limited diagnostic value. DWI reflects the free diffusion of water molecules in biologic tissue [27]. Because PCNSLs are highly cellular tumors with a high nuclear-to-cytoplasmic ratio, diffusion restriction has been reported as one of the important MRI findings [28]. Previous studies suggest that PCNSLs reveal more
AJR:204, April 2015 831
restricted diffusion and lower ADC values compared with high-grade gliomas [18, 19, 28]. In the study by Calli et al. [19], the ADCmin of PCNSL was reported as 510 ± 90 × 10 −6 mm2 /s, which helped to distinguish PCNSLs from other intracerebral malignancies. In our study, ADCmin and the low-percentile values determined from the cumulative ADC histogram (ADC5% and ADC10%) were found to be low in PCNSL, in which the ADCmin was 563 ± 90 ×10 −6 mm2 /s, which was in accordance with the results of previous literature reports. The typical DWI feature of TDLs is their variable ADC values [11, 14, 29]. It has been reported that most TDLs have elevated ADC values because of their vasogenic edema as well myelin destruction with axonal preservation [30]. Similar findings were also noted in our study. However, acute demyelinating lesions may show areas of reduced ADC values that are thought to be caused by intramyelinic edema or myelin vacuolation [31]. Contrast enhancement is considered a hallmark of biologically active lesions and reflects a breakdown in the blood-brain barrier [29]. In our TDL group, mostly very low ADCmin was observed in the enhancing portion of lesions in patients with the openring enhancement pattern, and peripheral restricted diffusion was the primary manifestation seen on DWI. Nevertheless, we found that the ADCmin of TDL was not as low as that of PCNSL and was therefore useful for discriminating PCNSLs because of its good diagnostic performance. Although the diagnostic performance of ADC5% and ADC10% does not seem to be as successful as that of ADCmin, the cumulative percentiles derived from a cumulative histogram may be more reliable parameters than the minimum values because they may be less affected by misregistration artifacts or imperfect delineation of lesion boundaries [20]. MRI features of atypical PCNSLs are more heterogeneous compared with homogeneously enhancing typical PCNSLs. In nine of the 16 atypical PCNSLs in our patient groups, it was not easy to delineate the tumor portions because they were obscured by edema. Therefore, in our study, we included all of the high-signal-intensity abnormalities seen on T2-weighted imaging for the ADC histogram analysis, including necrosis and edema, to obtain a more accurate evaluation of heterogeneous lesions. All PCNSLs showed higher SD than TDLs. We think that more severe ede-
832
A
B
C
D
160 140 120 Frequency
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
Lu et al.
100 80 60 40 20 0 600
800 ADCmin
1000 1200 1400 1600 1800 ADC5% ADC10% ADC Value (10−6 mm2/s)
2000
2200
2400
E Fig. 3—37-year-old woman with biopsy-proven tumefactive demyelinating lesion mimicking primary CNS lymphoma (PCNSL). A, T2-weighted image shows well-defined lesion in right frontal lobe with slightly high signal intensity and relatively little surrounding vasogenic edema. B, Contrast-enhanced T1-weighted image shows obvious patchy enhancement. C and D, Diffusion-weighted image (C) and corresponding apparent diffusion coefficient (ADC) map (D) reveal slightly decreased diffusion within lesion. E, ADC histogram based on entire lesion as well as edema shows wide distribution. Minimum ADC (ADCmin), 5th-percentile ADC (ADC 5%), and 10th-percentile ADC (ADC10%) values were 771, 1092, and 1198 × 10 −6 mm2 /s, respectively, all of which are higher than those of PCNSL.
AJR:204, April 2015
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
Histogram Analysis of ADC Maps ma, hemorrhage, and necrosis contribute to the greater variation of ADC values in PCNSLs. Other distribution parameters, including skewness and kurtosis, did not show statistically significant differences. Most TDLs (12/18) and atypical PCNSLs (13/16) showed positive skewness. Eighteen of 31 typical PCNSLs showed negative skewness, which may be due to the more severe edema in these cases. The low ADC values of the histogram may correspond to the densely packed lymphoid cells in PCNSLs or to the acute demyelinating plaque in TDLs, which usually appear as the enhancing portion of the lesions. The high ADC values are possibly associated with a mixture of viable and necrotic cells and edema. The sensitivity and specificity for ADCmin to differentiate all PCNSLs from TDLs were low (sensitivity, 68.1%; specificity, 77.8%). However, those to differentiate atypical PCNSLs from TDLs were high (sensitivity, 81.3%; specificity, 88.9%). It is unclear why there is a discrepancy between the two comparisons. We speculate that perfusion effect may have influenced the ADC value. Increased perfusion has been reported to cause increased ADC in the liver and also in brain tumors [25, 32]. Although perfusion in PCNSLs is not high, they still have increased perfusion. Solid-enhancing masses in typical PCNSLs may have higher perfusion than that of atypical PCNSLs, which usually have heterogeneous enhancement, and this may explain the higher ADCmin in the entire PCNSLs group than in the atypical PCNSLs group. Our findings suggest that histogram analysis of ADC is especially useful for differentiating atypical PCNSLs from TDLs, which is sometimes challenging in clinical practice. Nonenhancing PCNSL usually causes a diagnostic dilemma. In our previous study, we found that 1H-MR spectroscopy of nonenhancing PCNSLs was nonspecific and could not help in discriminating them from TDLs [26]. However, the ADCmin, ADC5%, and ADC10% of the nonenhancing PCNSLs in our series were lower than those of the TDLs, which indicates that ADC histogram analysis might be helpful in making this differentiation. However, because there was only one case in our study, further studies will be necessary to substantiate our findings. We found that ADC histograms were asymmetric, had broad distribution, and were bimodal or even multimodal, which agreed with previous reports [33]. Therefore,
we believe that the gaussian mixture modeling used in our study could more accurately provide better curve fitting and reflect the heterogeneity of the lesions in their entirety. There are several limitations to our study. First, the number of patients was relatively small and it was a retrospective study. More PCNSLs, especially atypical PCNSLs, and TDLs will be required in future studies to strengthen the statistical power. Second, accurately defining the boundary of lesions is challenging, especially when diffuse confluent lesions are shown. However, given the large number of pixel data included in the histogram of an entire lesion, the imperfect delineation of lesion boundaries is relatively unimportant according to the results of previous studies. Last, coregistration of T2weighted imaging with the ADC map was required in our study. Although motion correction had been performed and ROIs were carefully drawn, there is still the possibility of misregistration artifacts and the inclusion of erroneous ADC values at the lesion boundary. In that respect, cumulative histogram parameters, such as ADC5% and ADC10%, may be more promising. In conclusion, our results suggest that ADC histogram analysis based on an entire lesion could help to discriminate PCNSLs and TDLs and may be particularly useful for differentiating the subgroup of atypical PCNSLs from TDLs. The ADCmin and the low-percentile values of the cumulative ADC histogram, including ADC5% and ADC10%, may be potentially promising parameters. References 1. Küker W, Nägele T, Korfel A, et al. Primary central nervous system lymphomas (PCNSL): MRI features at presentation in 100 patients. J Neurooncol 2005; 72:169–177 2. Zacharia TT, Law M, Naidich TP, Leeds NE. Central nervous system lymphoma characterization by diffusion-weighted imaging and MR spectroscopy. J Neuroimaging 2008; 18:411–417 3. Haldorsen IS, Espeland A, Larsson EM. Central nervous system lymphoma: characteristic findings on traditional and advanced imaging. AJNR 2011; 32:984–992 4. Haldorsen IS, Kråkenes J, Krossnes BK, Mella O, Espeland A. CT and MR imaging features of primary central nervous system lymphoma in Norway, 1989–2003. AJNR 2009; 30:744–751 5. Tang YZ, Booth TC, Bhogal P, Malhotra A, Wilhelm T. Imaging of primary central nervous system lymphoma. Clin Radiol 2011; 66:768–777 6. Erdag N, Bhorade RM, Alberico RA, Yousuf N,
Patel MR. Primary lymphoma of the central nervous system: typical and atypical CT and MR imaging appearances. AJR 2001; 176:1319–1326 7. Jahnke K, Schilling A, Heidenreich J, et al. Radiologic morphology of low-grade primary central nervous system lymphoma in immunocompetent patients. AJNR 2005; 26:2446–2454 8. Lachenmayer ML, Blasius E, Niehusmann P, et al. Non-enhancing primary CNS lymphoma. J Neurooncol 2011; 101:343–344 9. Trendelenburg G, Zimmer C, Forschler A, Stadelmann C, Zschenderlein R. Atypical appearance of a primary central nervous system lymphoma. Arch Neurol 2006; 63:908–909 10. Ragel BT, Fassett DR, Baringer JR, Browd SR, Dailey AT. Decompressive hemicraniectomy for tumefactive demyelination with transtentorial herniation: observation. Surg Neurol 2006; 65:582–583 11. Given CA 2nd, Stevens BS, Lee C. The MRI appearance of tumefactive demyelinating lesions. AJR 2004; 182:195–199 12. Saindane AM, Cha S, Law M, Xue X, Knopp EA, Zagzag D. Proton MR spectroscopy of tumefactive demyelinating lesions. AJNR 2002; 23:1378–1386 13. Cha S, Pierce S, Knopp EA, et al. Dynamic contrast-enhanced T2*-weighted MR imaging of tumefactive demyelinating lesions. AJNR 2001; 22:1109–1116 14. Al-Okaili RN, Krejza J, Wang S, Woo JH, Melhem ER. Advanced MR imaging techniques in the diagnosis of intraaxial brain tumors in adults. RadioGraphics 2006; 26(suppl 1):S173–S189 15. Schroeder PC, Post MJ, Oschatz E, Stadler A, Bruce-Gregorios J, Thurnher MM. Analysis of the utility of diffusion-weighted MRI and apparent diffusion coefficient values in distinguishing central nervous system toxoplasmosis from lymphoma. Neuroradiology 2006; 48:715–720 16. Horger M, Fenchel M, Nägele T, et al. Water diffusivity: comparison of primary CNS lymphoma and astrocytic tumor infiltrating the corpus callosum. AJR 2009; 193:1384–1387 17. Camacho DL, Smith JK, Castillo M. Differentiation of toxoplasmosis and lymphoma in AIDS patients by using apparent diffusion coefficients. AJNR 2003; 24:633–637 18. Doskaliyev A, Yamasaki F, Ohtaki M, et al. Lymphomas and glioblastomas: differences in the apparent diffusion coefficient evaluated with high b-value diffusion-weighted magnetic resonance imaging at 3T. Eur J Radiol 2012; 81:339–344 19. Calli C, Kitis O, Yunten N, Yurtseven T, Islekel S, Akalin T. Perfusion and diffusion MR imaging in enhancing malignant cerebral tumors. Eur J Radiol 2006; 58:394–403 20. Kang Y, Choi SH, Kim YJ, et al. Gliomas: histogram analysis of apparent diffusion coefficient
AJR:204, April 2015 833
Downloaded from www.ajronline.org by UCSF LIB & CKM/RSCS MGMT on 04/18/15 from IP address 169.230.243.252. Copyright ARRS. For personal use only; all rights reserved
Lu et al. maps with standard- or high-b-value diffusionweighted MR imaging—correlation with tumor grade. Radiology 2011; 261:882–890 21. Nowosielski M, Recheis W, Goebel G, et al. ADC histograms predict response to anti-angiogenic therapy in patients with recurrent high-grade glioma. Neuroradiology 2011; 53:291–302 22. Al-Okaili RN, Krejza J, Woo JH, et al. Intraaxial brain masses: MR imaging-based diagnostic strategy—initial experience. Radiology 2007; 243:539–550 23. Toh CH, Wei KC, Ng SH, Wan YL, Castillo M, Lin CP. Differentiation of tumefactive demyelinating lesions from high-grade gliomas with the use of diffusion tensor imaging. AJNR 2012; 33:846–851 24. Yap KK, Sutherland T, Liew E, Tartaglia CJ, Pang M, Trost N. Magnetic resonance features of pri-
mary central nervous system lymphoma in the immunocompetent patient: a pictorial essay. J Med Imaging Radiat Oncol 2012; 56:179–186 25. Cohen AD, LaViolette PS, Prah M, et al. Effects of perfusion on diffusion changes in human brain tumors. J Magn Reson Imaging 2013; 38:868–875 26. Lu SS, Kim SJ, Kim HS, et al. Utility of proton MR spectroscopy for differentiating typical and atypical primary central nervous system lymphomas from tumefactive demyelinating lesions. AJNR 2014; 35:270–277 27. Koh DM, Collins DJ. Diffusion-weighted MRI in the body: applications and challenges in oncology. AJR 2007; 188:1622–1635 28. Guo AC, Cummings TJ, Dash RC, Provenzale JM. Lymphomas and high-grade astrocytomas: comparison of water diffusibility and histologic characteristics. Radiology 2002; 224:177–183
29. Balashov KE, Lindzen E. Acute demyelinating lesions with restricted diffusion in multiple sclerosis. Mult Scler 2012; 18:1745–1753 30. Tievsky AL, Ptak T, Farkas J. Investigation of apparent diffusion coefficient and diffusion tensor anisotropy in acute and chronic multiple sclerosis lesions. AJNR 1999; 20:1491–1499 31. Abou Zeid N, Pirko I, Erickson B, et al. Diffusionweighted imaging characteristics of biopsy-proven demyelinating brain lesions. Neurology 2012; 78:1655–1662 32. Hollingsworth KG, Lomas DJ. Influence of perfusion on hepatic MR diffusion measurement. NMR Biomed 2006; 19:231–235 33. Pope WB, Lai A, Mehta R, et al. Apparent diffusion coefficient histogram analysis stratifies progressionfree survival in newly diagnosed bevacizumab- treated glioblastoma. AJNR 2011; 32:882–889
F O R YO U R I N F O R M AT I O N
Unique customized medical search engine service from ARRS! ARRS GoldMiner® is a keyword- and concept-driven search engine that provides instant access to radiologic images published in peer-reviewed journals. ARRS members earn 0.5 CME credits for each self-directed search. For more information, visit goldminer.arrs.org.
834
AJR:204, April 2015
