Musculoskeletal Imaging Original Research

Transcription

1 Musculoskeletal Imaging Original Research Bonarelli et al. DWI Characterization of Nonfatty Soft-Tissue Lesions Musculoskeletal Imaging Original Research Chloé Bonarelli 1 Pedro Augusto Gondim Teixeira 1,2 Gabriela Hossu 1,2,3 Jean-Baptiste Meyer 1 Bailiang Chen 2,3 Frédérique Gay 4 Alain Blum 1 Bonarelli C, Gondim Teixeira PA, Hossu G, et al. Keywords: apparent diffusion coefficient (ADC) calculation, diffusion-weighted MRI, postprocessing method, soft tissues, tumoral characterization DOI: /AJR Received September 12, 2014; accepted after revision December 30, Service de Radiologie Guilloz, CHU Nancy, Rue du Maréchal de Lattre de Tassigny, Nancy, France. Address correspondence to C. Bonarelli (chloe. bonarelli@yahoo.fr). 2 CICIT, Vandoeuvre-lès-Nancy, France. 3 INSERM, Nancy, France. 4 CHUV Lausanne, Lausanne, Switzerland. WEB This is a web exclusive article. AJR 2015; 205:W106 W X/15/2051 W106 American Roentgen Ray Society Impact of ROI Positioning and Lesion Morphology on Apparent Diffusion Coefficient Analysis for the Differentiation Between Benign and Malignant Nonfatty Soft-Tissue Lesions OBJECTIVE. The objective of our study was to assess the impact of two methods of apparent diffusion coefficient (ADC) selection on the diagnostic performance of DWI in the characterization of nonfatty soft-tissue masses. SUBJECTS AND METHODS. Sixty-five histologically confirmed soft-tissue tumors imaged from November 2009 through October 2012 were evaluated. The minimal ADC ( ) and average ADC ( ) for each tumor were obtained using two ROI-positioning methods: manual and semiautomatic. Two readers correlated the findings to lesion histology and morphology on conventional MRI sequences. RESULTS. The obtained using the manual method presented a better sensitivity with a similar specificity when compared with the obtained using the semiautomatic method (manual vs semiautomatic: 75 83% and 59 63% vs 58 67% and 63% and 63%, respectively). The interobserver agreement for the was similar between the ADC selection methods (intraclass correlation coefficient = 0.81 and 0.87 for manual and semiautomatic methods, respectively). In the subgroup of solid lesions, the obtained using the manual method offered a better sensitivity for benign-malignant differentiation (60 81% vs 60% and 60% for and, respectively). CONCLUSION. The obtained with manual ROI positioning offered the best diagnostic performance for tumor characterization. The semiautomatic method yielded similar specificity. For solid masses, the were better correlated with tumor histology than the. M RI is routinely performed to characterize soft-tissue masses in various organs and systems. Although some lesions present specific morphologic and signal features that allow confident characterization (intratumoral fat, hemorrhage showing low signal intensity on T2-weighted imaging and high signal intensity on T1-weighted imaging), many soft-tissue tumors remain indeterminate after conventional MRI evaluation [1 4]. Furthermore, morphologic imaging is sometimes inadequate for the evaluation of treatment response [5]. Studies have shown that DWI with apparent diffusion coefficient (ADC) analysis can be useful for tumor characterization [6, 7]. Initial evaluations of DWI for differentiating benign from malignant masses yielded contradictory results depending on the histologic characteristics of the tumors being evaluated [7 10]. In clinical practice, the main impediment to tumor characterization with DWI is the marked histologic variability of tumors. This variability limits the correlation of ADC with tumor cellularity, a marker of tumor aggressiveness [6, 11, 12]. An optimal acquisition protocol is needed for ADC analysis. Many studies in the literature have assessed parameters for diffusion acquisitions in specific organs or in particular subgroups of tumors [13 18]. A consensus on DWI protocol parameters was established in 2009 [19]. Any given tumor will present a spectrum of ADC, and choosing one that is most representative of tumor behavior can be challenging. Variations in the size and position of the ROI can lead to significant changes in the ADC. It is unknown whether the morphologic aspect (e.g., cystic, completely solid, partially necrotic, and so on) of lesions should be considered when determining the optimal ROI position. The reproducibility of the ADC is also influenced by the type of postprocessing method. Semiautomatic methods of ADC calculation may facilitate the comparison of. The op- W106 AJR:205, July 2015

2 DWI Characterization of Nonfatty Soft-Tissue Lesions timal ADC selection method for clinical use and the impact it may have on the diagnostic performance of DWI for tumor characterization have not yet been determined. In this study, we assessed the impact of the ADC selection method on the diagnostic performance of DWI in the characterization of a histologically heterogeneous group of tumors (nonfatty soft-tissue masses) [20]. Average ADC ( ) and minimal ADC ( ) were selected using two simple and readily available ROI-positioning methods. Tumor morphology was correlated with the performance of the ADC selection method. Subjects and Methods Patients From November 2009 through October 2012, 400 patients with a suspected bone or soft-tissue neoplasm were prospectively included in a research protocol for the evaluation of noninvasive tumor characterization methods on MRI. This study was approved by the institutional ethics committee, and all patients were more than 18 years old and signed an informed consent form. Patients were referred for the initial lesion evaluation and had no history of treatment. DWI was performed in all cases. A total of 214 soft-tissue tumors were considered for this study. Sixty-three cases were excluded because no histologic confirmation was available (e.g., patients did not return for biopsy or refused biopsy or the imaging characteristics were sufficient for diagnosis), and four cases were excluded because the histologic results were inconclusive. Ten additional tumors were excluded because of significant motion artifacts on the DW images. Finally, 68 lipomatous tumors and four hemangiomas were also excluded because water diffusion in tumors containing fat differs considerably from that in nonfatty tumors, and thus these tumors are not directly comparable. Hence, the study population comprised 65 patients with histologically confirmed soft-tissue masses of the extremities or trunk. Pregnancy, a history of surgery, severe kidney failure, MRI contraindications, and patient refusal were additional exclusion criteria. Figure 1 shows the composition of the study population. MRI Examination MRI was performed using a 1.5-T unit (Signa HDxt, GE Healthcare) with dedicated coils. Conventional MRI sequences and DWI were performed during the same examination. The imaging protocol included at least one T1-weighted sequence, two T2-weighted fat-saturated fast spin-echo sequences in two different orthogonal planes, T1-weighted fatsaturated sequences after gadolinium injection, and Fig. 1 Flowchart shows process of selecting study group. 186 Bone tumors DWI. The acquisition parameters were adapted to the anatomic region being assessed. DWI was performed before contrast injection. A single-shot pulsed gradient spin-echo DWI sequence with echo-planar imaging readout was performed at all three imaging axis directions. A single b value setting was used with a b value equal to 600 s/mm 2 after one non-dwi measurement (b = 0 s/mm 2 ). The other acquisition parameters were as follows: TR, 5000 ms; TE, minimal; number of signals acquired, 6; bandwidth, 250 Hz; slice thickness, 6 mm; gap, 0; and acquisition matrix, The FOV and the slice thickness were adapted to the patient s anatomy. The acquisition time of the DWI sequence was 1 minute 40 seconds. Image Analysis All images were transferred to a workstation console (Advantage Windows, version 4.4, GE Healthcare) using the FuncTool application (version 4.4, GE Healthcare), which was used for ADC calculation, functional diffusion map creation, and ROI positioning. The images were analyzed by two radiologists with 3 and 8 years of clinical experience using MRI; they interpreted the images independently and were blinded to the clinical and histologic information. The readers classified the tumors as solid, cystic, or mixed (cystic and solid) on the basis of their morphologic features on contrast-enhanced T1-weighted fat-saturated and T2-weighted fat-saturated images. The and were calculated using two different ROI-positioning methods: manual and semiautomatic (Fig. 2). Manual method Functional diffusion maps were correlated with axial T2-weighted fat-saturated images and contrast-enhanced T1-weighted fat-saturated images to allow identification of areas of tumor necrosis. All slices depicting tumor were analyzed. First, an elliptical ROI was placed on the nonnecrotic tumor area that presented the lowest ADC value. 67 Soft-tissue masses with no histologic results or inconclusive histologic results 10 Tumors with significant motion artifacts on DWI 400 Patients evaluated from November 2011 through October Soft-tissue masses 72 Fatty or vascular tumors 147 Histologically confirmed softtissue tumors 65 Nonfatty histologically confirmed softtissue tumors (study population) The size of the ROI was determined by the reader. The mean ADC value of the pixels contained in this ROI was considered to represent the of the tumor. Areas of flow void (intratumoral tubular images of signal void), calcification (amorphous hypointense foci in all sequences), and normal tissue were avoided during ROI placement. Then, the image containing the largest tumor diameter was selected and an elliptical ROI was placed to include the largest portion possible of the mass. The mean value of the pixels contained in this ROI was considered to represent the of the tumor. Semiautomatic method The image containing the largest portion of the tumor was selected. A free-form ROI was drawn over the tumor borders. The and of this ROI were calculated with an automatic pixel-by-pixel analysis. With this method, the were related to the pixel size and the slice thickness. TABLE 1: Histologic Subtypes of the 65 Tumors in the Study Population Histologic Subtype No. of Tumors Sarcomas 23 Giant cell tumors 10 Schwannomas 7 Myxoid tumors 6 Desmoid tumors 4 Carcinomas 4 Leiomyomas 2 Villonodular synovitis 2 Fibromatosis 2 Synovial chondromatosis 2 Glomic tumor 1 Melanoma 1 Histiocytofibroma 1 AJR:205, July 2015 W107

3 Bonarelli et al. These ROI-positioning procedures were performed by both readers independently with an interval of more than 2 weeks between the manual and the semiautomatic readouts. The resulting four ADC value datasets were correlated to lesion morphology and histopathologic findings. The maximum ADC was not considered in the analysis because it generally represents areas of myxoid, cystic, or necrotic tissue and is poorly correlated with tumor behavior [21]. Statistical Analysis The unpaired Student t test was used to determine the significance of the differences in ADC between benign and malignant lesions and among the different tumor morphologic types. A D E Fig year-old woman who presented for MRI, including DWI, of high-grade sarcoma of arm. A, Axial T2-weighted fat-saturated MR image shows large hyperintense subaponeurotic soft-tissue mass (arrows). Cystic or necrotic areas (arrowheads) are seen within lesion. B, Axial T1-weighted fat-saturated gadolinium-enhanced MR image shows prominent nonenhancing necrotic areas. Nodular zone of viable tumoral tissue (arrow) is seen within mass. C, DW image shows selection of minimal apparent diffusion coefficient ( ) using manual method for ROI positioning. After DW images were correlated with conventional MR images, ROI (circle, 1) was positioned on nonnecrotic tumor area with lowest ADC available. D, DW image shows selection of average apparent diffusion coefficient ( ) using manual method for ROI positioning. On image containing larger tumor diameter, ROI was placed to occupy as much tumor as possible (oval, calipers). Note that ROI size is dependent on lesion diameter and that ROI used for selection is larger than ROI used for selection. E, DW image shows semiautomatic method of ADC selection. On image containing larger tumor diameter, ROI was drawn over tumor margins (pink free-form outline, 1). and (not shown) are then provided automatically by software using pixel-by-pixel analysis. Note that size of ROI used for selection (C) is larger than pixels of this image. For all analyses, p < 0.05 was considered as the threshold for statistical significance. The intraclass correlation coefficient (ICC) was used to evaluate the reproducibility of the ROI selection methods. An ROC analysis was performed for each ADC dataset. On the basis of the ROC analysis, the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy of benign-malignant differentiation were calculated. A linear regression analysis was used to study ADC variations related to lesion size, localization, and morphology. Statistical analysis was performed using statistics software (R, version 3.01, R Project for Statistical Computing, including its proc Library [22]. B Results The study population comprised 35 men and 30 women (male-female ratio, 35:30 [1.17]) with an age range of years (mean, 50 years). After histologic analysis, 41 benign and 24 malignant tumors were identified. Their histologic subtypes are presented in Table 1. The mean tumor diameters were 93 ± 43 (SD) mm for the malignant tumors and 42 ± 33 mm for the benign lesions. The anatomic distribution was as follows: 40 lesions in the lower limb, 21 in the upper limb, two in the chest wall, one in the head and neck region, and one in the abdominal wall. The mean area of the ROI used for manual analysis was 72.8 ± mm 2 for C W108 AJR:205, July 2015

4 DWI Characterization of Nonfatty Soft-Tissue Lesions Sensitivity Specificity Manual Manual Semiautomatic Semiautomatic Fig. 3 ROC analysis of diagnostic performance of apparent diffusion coefficient (ADC) datasets studied for differentiation between benign and malignant nonfatty soft-tissue tumors. = minimal ADC, = average ADC. reader 1 and 90.1 ± mm 2 for reader 2. The mean area of the ROI used for manual analysis was ± mm 2 for reader 1 and ± mm 2 for reader 2. Finally, the mean area of the ROI used for semiautomatic analysis was ± mm 2 for reader 1 and ± 2154 mm 2 for reader 2. Because the with the semiautomatic method were obtained using a pixel-by-pixel analysis, the ROI size equals that of the pixel size. There was a statistically significant difference between the ADC of benign and malignant tumors in all the ADC datasets (p < 0.01). When all lesions were considered (benign and malignant), the obtained with the manual and semiautomatic ROI selection methods were statistically significantly different (p < ). However, there were no significant differences in the regardless of the ROI selection method (p = 0.71). The reproducibility of the ROI selection methods was considered excellent for and good for with both ROI selection methods. The ICC and mean of the four ADC datasets in benign and malignant tumors are presented in Table 2. Linear regression showed a significant interaction of with tumor histologic nature (benign vs malignant) independent of the ROI selection method (p = 0.02). The tumor histologic nature did not significantly influence (p > 0.11). Linear regression showed lesion size and location did not influence the calculated ADC (p > 0.05). Histologic results were used as the reference standard, and ROC curves were constructed for the available ADC datasets. The AUC was 0.73 for and 0.59 for obtained using the manual method, and the AUC was 0.68 for the and 0.62 for obtained using the semiautomatic method (Fig. 3). The best sensitivity, specificity, PPV, NPV, and accuracy for each of the ADC datasets under evaluation for the differentiation between benign and malignant lesions and the used cutoff are presented in Table 3. The manual showed the best benign-malignant differentiation with a sensitivity, specificity, PPV, NPV, and accuracy for both readers of 75 83%, 59 63%, 52 57%, 80 87%, and 65 70%, respectively. The semiautomatic presented an overall lower performance (sensitivity, 58 67%) and similar specificity (63 63%). The performance of the was inferior regardless of the ROI selection method. For the, the manual ROI selection method yielded a sensitivity of 58 63% and specificity of 51 57% for benign-malignant differentiation, whereas semiautomatic ROI positioning yielded a sensitivity of 58 63% and specificity of 56% and 56%. The PPV, NPV, and accuracy of and with semiautomatic ROI positioning were similar (44 45%, 70 72%, and 57 58% vs 48 52%, 72 77%, and 62 65%, respectively). When tumors were classified according to morphology, six were cystic (all benign), 21 were solid (16 benign and five malignant), and 38 were mixed (19 benign and 19 malignant). The mean and SD of ADC in these subgroups are presented in Table 4. The performance of the different ADC datasets was calculated for solid and mixed tumors using the same cutoffs (Table 5). Tumor morphology had an influence on the diagnostic performance of the ADC for benign-malignant differentiation. When solid tumors were considered, the diagnostic performance of manual declined considerably (sensitivity, specificity, PPV, and NPV for both readers: 60% and 60%, 50 58%, 26 27%, and 80 83%, respectively). In this group, however, manual performed better than for both readers, with a higher sensitivity (60 81%), similar specificity (56 69%), and slightly better NPV (82 85%); the PPV was 30 37%. With the semiautomatic method, there were no noticeable performance differences with the and. When mixed lesions were considered, the performed better than the regardless of the ROI selection method. The specificity of the was better with the semiautomatic method (68 79%) than with the manual method (63 68%) (Table 5). Discussion On DWI, tumors present a spectrum of ADC ; the ADC are related to the intratumor regional histologic character- TABLE 2: Apparent Diffusion Coefficient (ADC) Values Obtained in Benign and Malignant Tumors Using Different ROI-Positioning Methods ADC Values Benign Malignant ICC Manual method, mean ± SD ( 10 3 mm 2 /s) 0.81 Reader ± ± 0.26 Reader ± ± 0.40, mean ± SD ( 10 3 mm 2 /s) 0.76 Reader ± ± 0.47 Reader ± ± 0.48 Semiautomatic method, mean ± SD ( 10 3 mm 2 /s) 0.87 Reader ± ± 0.47 Reader ± ± 0.40, mean ± SD ( 10 3 mm 2 /s) 0.71 Reader ± ± 0.52 Reader ± ± 0.49 Note ICC = intraclass correlation coefficient, = minimal ADC, = average ADC. AJR:205, July 2015 W109

5 TABLE 3: Diagnostic Performance of Apparent Diffusion Coefficient (ADC) Values for Differentiation Between Benign and Malignant Lesions Using Different Bonarelli ROI-Positioning et al. Methods Manual method (cutoff = mm 2 /s) ADC Cutoffs Reader 1 Reader 2 Sensitivity (%) Specificity (%) PPV (%) NPV (%) Accuracy (%) (cutoff = mm 2 /s) Sensitivity (%) Specificity (%) PPV (%) NPV (%) Accuracy (%) Semiautomatic method (cutoff = mm 2 /s) Sensitivity (%) Specificity (%) PPV (%) NPV (%) Accuracy (%) (cutoff = mm 2 /s) Sensitivity (%) Specificity (%) PPV (%) NPV (%) Accuracy (%) Note = minimal ADC, PPV = positive predictive value, NPV = negative predictive value, = average ADC. TABLE 4: Apparent Diffusion Coefficient (ADC) Values Obtained in the Different Morphologic Tumor Subgroups Using Different ROI-Positioning Methods Manual method, mean ± SD ( 10 3 mm 2 /s) ADC Value Cystic Lesions Solid Lesions Mixed Lesions Reader ± ± ± 0.59 Reader ± ± ± 0.61, mean ± SD ( 10 3 mm 2 /s) Reader ± ± ± 0.55 Reader ± ± ± 0.61 Semiautomatic method, mean ± SD ( 10 3 mm 2 /s) Reader ± ± ± 0.54 Reader ± ± ± 0.46, mean ± SD ( 10 3 mm 2 /s) Reader ± ± ± 0.56 Reader ± ± ± 0.49 Note = minimal ADC, = average ADC. W110 AJR:205, July 2015

6 DWI Characterization of Nonfatty Soft-Tissue Lesions TABLE 5: Diagnostic Performance of Apparent Diffusion Coefficient (ADC) Values in Two Tumor Morphologic Subgroups Manual Method Semiautomatic Method Performance Value Reader 1 Reader 2 Reader 1 Reader 2 Reader 1 Reader 2 Reader 1 Reader 2 Solid tumors Sensitivity (%) Specificity (%) PPV (%) NPV (%) Mixed tumors Sensitivity (%) Specificity (%) PPV (%) NPV (%) Note = minimal ADC, = average ADC, PPV = positive predictive value, NPV = negative predictive value. istics (i.e., cellularity, tissue organization, and morphology of the interstitial space) [19]. We used the same acquisition protocol but found a significant variation in the ADC depending on the method used for ROI positioning (p < 0.01). Similar findings have been reported by Lambregts et al. [23] in the follow-up of patients with rectal cancer treated with chemotherapy and radiotherapy. The results presented here indicate that ADC value selection has a nonnegligible impact on the performance of DWI for benignmalignant differentiation. Manual ROI selection based on the correlation between functional diffusion maps and morphologic images (T2-weighted and gadolinium-enhanced T1-weighted sequences) is time-consuming and more subjective than semiautomatic ROI selection. Nevertheless, the interobserver variability of the manual ADC selection method evaluated here was considered excellent for and good for (ICC = 0.81 and 0.76, respectively). The use of a semiautomatic ROI selection method, which tends to be less subjective, led to a slight increase in the ICC for (ICC = 0.87). With semiautomatic ROI positioning, the reproducibility was reduced (ICC = 0.71), probably because of difficulties in the precise determination of tumor margins on functional diffusion maps. The ADC obtained using the manual ROI selection showed better diagnostic performance than the ADC obtained using the semiautomatic method. The use of, regardless of the postprocessing method, was associated with a better diagnostic performance overall. The obtained using manual ROI selection performed better than those obtained using semiautomatic ROI selection. The sensitivity and NPV of the were 75 83% and 80 87% with the manual method versus 58 67% and 72 77% with the semiautomatic method, respectively. Despite the lower sensitivity with the semiautomatic ROI selection, the specificity was similar (59 63% for the manual method and 63% and 63% for the semiautomatic method) and the reproducibility was slightly better. These results in addition to the fact that semiautomatic selection is less time-consuming make this method likely to be the most suited for initial tumor characterization. The tend to reflect global tumor histology including myxoid and necrotic areas, which are less useful for characterization but may be useful for evaluation of treatment response [23]. The correlation of ADC with lesion morphology has shown that better represent tumor histology for homogeneously enhancing solid masses with no detectable macroscopic necrotic or myxoid predominant areas. In this group of 21 lesions, the analysis presented a sensitivity, specificity, and NPV that varied from 60% to 81%, 56% to 69%, and 82% to 85%, respectively. These findings suggest that a larger ROI including the tumor zone with different diffusion environments may be more representative of tumor behavior in homogeneously enhancing solid lesions. In accordance with studies in the literature, when mixed lesions were considered, the performed better than the [5, 9, 24]. Again, despite a lower sensitivity, the best specificity for mixed lesions were obtained with semiautomatic (68 79%). When tumor morphology was considered, DWI presented an NPV for malignancy that varied from 71% to 92%, which could be useful in tumor characterization. These results add to the evidence that the correlation of DWI findings with morphologic tumor features is important for selecting representative ADC. In accordance with a study in the literature [23], our results showed there was a relationship between the ADC and the ROI size in the datasets. The smaller ROIs used for selection compared with those used for selection yielded lower ADC, which had a higher overall diagnostic performance. However, when pixelby-pixel analyses were performed (ROI size corresponding with pixel size), a decrease in sensitivity was noted. This finding is probably related to the inclusion of pixels reflecting tumor components (e.g., flow voids, calcification, or densely fibrotic areas), which seem not to correlate well with tumor behavior. There is little information in the literature about the diagnostic performance of DWI in the characterization of soft-tissue tumors in adults. Although an excellent 100% specificity was achieved in the differentiation between cystic and solid soft-tissue lesions, the evaluation of solid and mixed tumors is not as straightforward [25]. In this group of tumors with a wide variety of histologic types, when ADC selection was adapted to the lesion morphology, the diagnostic performance was only slightly inferior to that reported for the differentiation between retroperitoneal fibrosis and malignant tumors [26]. DWI seems to perform AJR:205, July 2015 W111

7 Bonarelli et al. better for the evaluation of pediatric soft-tissue tumors, with reported sensitivity and specificity of more than 90% [27]. The great histologic diversity of soft-tissue tumors in adults may account for these differences [20]. More complex methods of ADC selection that is, with volumetric analysis, global pixelby-pixel evaluation, and even histogram-based approaches with spatial tagging have been described in the literature [28, 29]. Although these techniques can improve the reproducibility of ADC measurements, they are timeconsuming, are not readily available, and are particularly suited to treatment response monitoring and evaluation [19, 23]. The ADC selection techniques evaluated here are easy to apply and have minimal software requirements [11, 30]. Similar ADC selection methods have been used previously, but there is limited information on ROI-positioning methods and the reproducibility of ADC measurements in tumors with a variety of histologic features. The results presented here may help define standards for DWI quantitative analysis in clinical practice. Several limitations to this study must be acknowledged. There was a high frequency of malignant lesions in the studied population, which is probably related to a selection bias. Our institution is a tertiary reference center for the evaluation of soft-tissue masses. However, this bias probably has little effect on the interpretation of the data presented because this study compares the variations in diagnostic performance of two ADC selection methods in the same population. A DWI protocol with three b is currently recommended in the literature [8]; however, this study was initiated in 2009 before these recommendations were made. The b chosen are within the range of the current recommendations [8, 19]. Although a dual b value protocol might lead to a lower diagnostic performance, it does not influence our results because we compared the diagnostic performance of the same protocol with different ADC selection methods. The number of lesions in the solid tumor group was relatively small, so these findings should be confirmed in larger patient populations. Follow-up examinations were not available, thus precluding evaluation of the intrapatient reproducibility of the ADC selection methods. Soft-tissue tumors can have a variety of histologic features, and the techniques that we studied must be evaluated in tumors of other organs and systems, which may present distinct histologic features. The DWI features of fatty lesions were not evaluated in this study. In conclusion, ADC selection influences ROI size and accounts for a significant variation in the diagnostic performance of DWI for benignmalignant differentiation of soft-tissue tumors. obtained with manual ROI positioning offered the best diagnostic performance. Compared with manual ROI positioning, semiautomatic ROI positioning yielded similar specificity, was more reproducible, and was less time-consuming; these advantages make the semiautomatic ROI-positioning method the best suited for initial tumor characterization. When lesion morphology was considered, the diagnostic performance of the semiautomatic method was improved, and a NPV of 71 92% was reached for solid and mixed lesions. Although should be used preferentially, may be better correlated with tumor histology in homogeneously enhancing solid masses. This knowledge could improve the analysis of quantitative DWI data in clinical practice. References 1. Gielen JL, De Schepper AM, Vanhoenacker F, et al. Accuracy of MRI in characterization of soft tissue tumors and tumor-like lesions: a prospective study in 548 patients. Eur Radiol 2004; 14: Chen CK, Wu HT, Chiou HJ, et al. Differentiating benign and malignant soft tissue masses by magnetic resonance imaging: role of tissue component analysis. J Chin Med Assoc 2009; 72: Costa FM, Canella C, Gasparetto E. Advanced magnetic resonance imaging techniques in the evaluation of musculoskeletal tumors. Radiol Clin North Am 2011; 49: , vii viii 4. Mayerhoefer ME, Breitenseher M, Amann G, Dominkus M. Are signal intensity and homogeneity useful parameters for distinguishing between benign and malignant soft tissue masses on MR images? Objective evaluation by means of texture analysis. Magn Reson Imaging 2008; 26: Oka K, Yakushiji T, Sato H, Hirai T, Yamashita Y, Mizuta H. The value of diffusion-weighted imaging for monitoring the chemotherapeutic response of osteosarcoma: a comparison between average apparent diffusion coefficient and minimum apparent diffusion coefficient. Skeletal Radiol 2010; 39: Einarsdóttir H, Karlsson M, Wejde J, Bauer HCF. Diffusion-weighted MRI of soft tissue tumours. Eur Radiol 2004; 14: Genovese E, Canì A, Rizzo S, Angeretti MG, Leonardi A, Fugazzola C. Comparison between MRI with spin-echo echo-planar diffusionweighted sequence (DWI) and histology in the diagnosis of soft-tissue tumours. Radiol Med (Torino) 2011; 116: Fayad LM, Jacobs MA, Wang X, Carrino JA, Bluemke DA. Musculoskeletal tumors: how to use anatomic, functional, and metabolic MR techniques. Radiology 2012; 265: Oka K, Yakushiji T, Sato H, et al. Usefulness of diffusion-weighted imaging for differentiating between desmoid tumors and malignant soft tissue tumors. J Magn Reson Imaging 2011; 33: Maeda M, Matsumine A, Kato H, et al. Soft-tissue tumors evaluated by line-scan diffusion-weighted imaging: influence of myxoid matrix on the apparent diffusion coefficient. J Magn Reson Imaging 2007; 25: van Rijswijk CSP, Kunz P, Hogendoorn PC, Taminiau AH, Doornbos J, Bloem JL. Diffusionweighted MRI in the characterization of soft-tissue tumors. J Magn Reson Imaging 2002; 15: Humphries PD, Sebire NJ, Siegel MJ, Olsen ØE. Tumors in pediatric patients at diffusion-weighted MR imaging: apparent diffusion coefficient and tumor cellularity. Radiology 2007; 245: Wu LM, Xu JR, Ye YQ, Lu Q, Hu JN. The clinical value of diffusion-weighted imaging in combination with T2-weighted imaging in diagnosing prostate carcinoma: a systematic review and meta-analysis. AJR 2012; 199: Doğanay S, Kocakoç E, Ciçekçi M, Ağlamiş S, Akpolat N, Orhan I. Ability and utility of diffusion-weighted MRI with different b in the evaluation of benign and malignant renal lesions. Clin Radiol 2011; 66: Watanabe Y, Yamasaki F, Kajiwara Y, et al. Preoperative histological grading of meningiomas using apparent diffusion coefficient at 3T MRI. Eur J Radiol 2013; 82: Mutlu H, Sivrioglu AK, Sonmez G, et al. Role of apparent diffusion coefficient and diffusion-weighted magnetic resonance imaging in differentiation between benign and malignant thyroid nodules. Clin Imaging 2012; 36: Thoeny HC, De Keyzer F, King AD. Diffusionweighted MR imaging in the head and neck. Radiology 2012; 263: Bogner W, Gruber S, Pinker K, et al. Diffusionweighted MR for differentiation of breast lesions at 3.0 T: how does selection of diffusion protocols affect diagnosis? Radiology 2009; 253: Padhani AR, Liu G, Koh DM, et al. Diffusionweighted magnetic resonance imaging as a cancer biomarker: consensus and recommendations. Neoplasia 2009; 11: Jo VY, Fletcher CD. WHO classification of soft tissue tumours: an update based on the 2013 (4th) edition. Pathology 2014; 46: Hirano M, Satake H, Ishigaki S, Ikeda M, Kawai H, Naganawa S. Diffusion-weighted imaging of breast masses: comparison of diagnostic performance using various apparent diffusion coeffi- W112 AJR:205, July 2015

8 DWI Characterization of Nonfatty Soft-Tissue Lesions cient parameters. AJR 2012; 198: MA, Fayad LM. Characterization of soft tissue 28. Yankeelov TE, Lepage M, Chakravarthy A, et al. 22. The Comprehensive R Archive Network website. masses: can quantitative diffusion weighted im- Integration of quantitative DCE-MRI and ADC Hornik K. R FAQ: frequently asked questions on aging reliably distinguish cysts from solid mass- mapping to monitor treatment response in human R. CRAN.R-project.org/doc/FAQ/R-FAQ.html. es? Skeletal Radiol 2013; 42: breast cancer: initial results. Magn Reson Imaging Published Accessed July Bakir B, Yilmaz F, Turkay R, et al. Role of diffu- 2007; 25: Lambregts DM, Beets GL, Maas M, et al. Tumour ADC measurements in rectal cancer: effect of ROI methods on ADC and interobserver variability. Eur Radiol 2011; 21: Nagata S, Nishimura H, Uchida M, et al. Diffusionweighted imaging of soft tissue tumors: usefulness of the apparent diffusion coefficient for differential diagnosis. Radiat Med 2008; 26: Subhawong TK, Durand DJ, Thawait GK, Jacobs sion-weighted MR imaging in the differentiation of benign retroperitoneal fibrosis from malignant neoplasm: preliminary study. Radiology 2014; 272: Neubauer H, Evangelista L, Hassold N, et al. Diffusion-weighted MRI for detection and differentiation of musculoskeletal tumorous and tumorlike lesions in pediatric patients. World J Pediatr 2012; 8: Moffat BA, Chenevert TL, Lawrence TS, et al. Functional diffusion map: a noninvasive MRI biomarker for early stratification of clinical brain tumor response. Proc Natl Acad Sci U S A 2005; 102: Herneth AM. Diffusion weighted imaging: have we found the Holy Grail of diagnostic imaging or is it still a game of numbers? Eur J Radiol 2003; 45: AJR:205, July 2015 W113