This copy is for personal use only. To order printed copies, contact Purpose: Materials and Methods: Results: Conclusion:

Transcription

1 This copy is for personal use only. To order printed copies, contact Original Research n Gastrointestinal Imaging Quantitative Assessment of Rectal Cancer Response to Neoadjuvant Combined Chemotherapy and Radiation Therapy: Comparison of Three Methods of Positioning Region of Interest for ADC Measurements at Diffusion-weighted MR Imaging 1 Ivana M. Blazic, MD, PhD Gordana B. Lilic, MD Milan M. Gajic, PhD Purpose: Materials and Methods: To determine the impact of three different methods of region of interest (ROI) positioning for apparent diffusion coefficient (ADC) measurements on the assessment of complete response (CR) to neoadjuvant combined chemotherapy and radiation therapy (CRT) in patients with rectal cancer. Institutional review board approval was obtained for this study; all patients gave written informed consent. ADCs were measured by two radiologists using three circular ROIs (three-rois), single-section (SS), and whole-tumor volume (WTV) methods in 62 patients with locally advanced rectal cancer on pre- and post-crt images. Interobserver variability was analyzed by calculating intraclass correlation coefficient (ICC). Descriptive statistics and areas under the receiver operating characteristic curves (AUCs) were calculated to evaluate performance in determining CR from pre- and post-crt ADCs and ADC change. Histopathologic tumor regression grade was the reference standard. An earlier incorrect version of this article appeared online. This article was corrected on September 19, From the Department of Radiology, Memorial Sloan- Kettering Cancer Center, 1275 York Ave, New York, NY (I.M.B.); Center for Radiology and MRI, Clinical Center of Serbia, Belgrade, Serbia (G.B.L.); and Institute for Medical Statistics and Informatics, Belgrade, Serbia (M.M.G.). Received August 28, 2015; revision requested October 19; final revision received January 19, 2016; accepted February 16; final version accepted March 26. Address correspondence to I.M.B. ( ivanablazic@yahoo.com). q RSNA, 2016 Results: Conclusion: SS and WTV methods yielded higher AUCs than did the three-rois method when determining CR from post-crt ADC (0.874 [95% confidence interval {CI}: 0.778, 0.970] and [95% CI: 0.781, 0.990] vs [95% CI: 0.583, 0.878], respectively; P =.033 and P =.003) and numeric change (0.892 [95% CI: 0.812, 0.972] and [95% CI: 0.801, 0.994] vs [95% CI: 0.591, 0.890], respectively; P =.048 and P =.0021). Respective accuracies of SS, WTV, and three-rois methods were 79% (49 of 62), 77% (48 of 62), and 61% (38 of 62) for post-crt, 79% (49 of 62), 86% (53 of 62), and 60% (37 of 62) for numeric ADC change, and 77% (48 of 62), 84% (52 of 62), and 57% (35 of 62) for percentage ADC change (ADC cut-offs: 1.21, 1.30, and mm 2 /sec, 0.33, 0.45, and mm 2 /sec increases, and 40%, 54%, and 27% increases, respectively). Post-CRT and ADC change measurements achieved negative predictive values of 96% (44 of 46) to 100% (39 of 39). Intraobserver agreement was highest for WTV-derived ADCs (ICC, [95% CI: 0.316, 0.892] to [95% CI: 0.615, 0.956]) and higher for all pretreatment than posttreatment measurements (ICC, [95% CI: 0.209, 0.930] and [95% CI: 0.164, 0.895] for three-rois method, [95% CI: 0.287, 0.844] and [95% CI: 0.176, 0.870] for SS method, [95% CI: 0.615, 0.956] and for WTV method [95% CI: 0.316, 0.892]). Tumor ADCs are highly dependent on the ROI positioning method used. Larger area measurements yield greater accuracy in response assessment. Post-CRT ADCs and values of ADC changes accurately identify noncomplete responders. WTV measurement of percentage ADC change provides the best results. q RSNA, radiology.rsna.org n Radiology: Volume 282: Number 2 February 2017

2 The treatment of rectal cancer has shifted from surgery followed by combined chemotherapy and radiation therapy (CRT) toward neoadjuvant CRT, with the choice of further Advances in Knowledge Variations in region of interest (ROI) number, size, and position substantially influence apparent diffusion coefficient (ADC) measurements; tumor ADCs are highly dependent on the particular ROI positioning method used (P = in the majority of comparisons). Post-combined chemotherapy and radiation therapy (CRT) ADC measurements and measurements of numeric ADC change covering all tumor areas, including areas of fibrosis and necrosis (on one or all sections), are more accurate for identifying complete tumor response to therapy than are measurements that include only viable tumor regions (accuracy, 77% 86% vs 60% 61%; area under the curve [AUC], to vs to 0.740; P = ). Whole-tumor volume (WTV) ADC measurements achieve higher reproducibility than measurements confined to tumor on a single section or to areas of viable tumor (intraclass correlation coefficient, vs ). Pretreatment ADC measurements are not reliable for assessing complete tumor response, regardless of the ROI positioning method (AUC, to 0.642). Measurements of ADC change induced by CRT may have considerable diagnostic value for determining complete tumor response, regardless of the ROI positioning method used (AUC, to 0.897); percentage of ADC change measured with WTV method provides the best diagnostic performance (AUC, 0.882; accuracy, 84%; sensitivity, 80%; specificity, 85%). therapy depending on the degree of response to CRT. The treatment should be adjusted to each patient based on the individual risk for local recurrence, and less invasive treatment strategies may be applied instead of standard rectal cancer surgery for tumors with good or complete response to CRT (1 4). Therefore, imaging evaluation of tumor response plays a crucial role in patient care as it may alter the treatment plaing (5,6). Different imaging techniques can be used to estimate tumor viability, cellularity, and vascularization, such as perfusion computed tomography (CT), fluorine 18 fluorodeoxyglucose positron emission tomography, and diffusionweighted (DW) magnetic resonance (MR) imaging. They offer the ability to evaluate response of rectal cancer to neoadjuvant CRT by combining quantitative assessment of tumor viability with qualitative assessment (7 11). DW MR imaging provides both excellent visual definition of tumor tissue due to high signal intensity and Implications for Patient Care Regardless of which ROI positioning method is used to obtain ADC measurements, post-crt ADCs and ADC changes both allow discrimination of noncomplete responders and thus accurate selection of patients with locally advanced rectal cancer who require surgical treatment (negative predictive value, 96% to 100%). Diagnostic performance levels in the assessment of complete tumor response to therapy were comparable with those of singlesection and WTV methods and for post-crt measurements and ADC change measurements; because acquiring post-crt measurements with the single-section method is a relatively efficient approach, it may be the best option for clinical practice (for this approach, AUC = 0.874, accuracy = 79%, sensitivity = 90%, specificity = 77%). quantitative information reflecting tissue cellularity. The apparent diffusion coefficient (ADC) has become an established quantitative biomarker for tumor characterization, as well as for prediction and monitoring of therapeutic response. Quantitative ADC data correlate highly with histologic findings and may indicate tumor viability (12 16). DW MR imaging enables evaluation of tumor response to CRT, but the best technique for performing measurements is still controversial, since it is unclear whether the entire tumor volume or representative parts of tumor tissue should be included in the analysis and whether the analysis of representative sections can reliably determine the level of response to therapy. Quantitative assessment of response to therapy by using DW MR imaging is performed by placing one or more regions of interest (ROIs) in the area of the tumor and measuring ADC values within the ROI(s). The tumor tissue is heterogeneous and often contains areas of necrosis, so the crucial questions are what size the ROI(s) should be, how many are needed, and where the ROI(s) should be positioned. Goh et al found that ROI placement substantially affected the ultimate values of Published online before print /radiol Content code: Radiology 2017; 282: Abbreviations: ADC = apparent diffusion coefficient AUC = area under the ROC curve CI = confidence interval CRT = combined chemotherapy and radiation therapy DW = diffusion weighted ROC = receiver operating characteristic ROI = region of interest Author contributions: Guarantors of integrity of entire study, I.M.B., G.B.L.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of final version of submitted manuscript, all authors; agrees to ensure any questions related to the work are appropriately resolved, all authors; literature research, I.M.B.; clinical studies, I.M.B., G.B.L.; statistical analysis, M.M.G.; and manuscript editing, I.M.B., G.B.L. Conflicts of interest are listed at the end of this article. Radiology: Volume 282: Number 2 February 2017 n radiology.rsna.org 419

3 colorectal cancer vascular parameters at CT (17). Lambregts et al discovered that variation in ROI size and positioning considerably influenced values of tumor ADC measurements (18). The purpose of our study was to determine the impact of three different methods of ROI positioning for ADC measurements on the assessment of complete response to neoadjuvant CRT in patients with rectal cancer. Materials and Methods Patients Eighty-two adult patients with locally advanced rectal cancer diagnosed at our institution between January 9, 2012, and January 31, 2014, enrolled in our prospective study. Institutional review board approval was obtained for this study, and all patients gave written informed consent. Inclusion criteria were (a) biopsyproven rectal adenocarcinoma; (b) locally advanced disease diagnosed at pre-crt T2-weighted MR imaging as stage ct3/t4 or as distal rectal cancer less than 5 mm from the anorectal junction and/or positive nodal stage (one or more lymph nodes larger than 5 mm and/or exhibiting heterogeneous signal intensity or an irregular border); (c) long-course neoadjuvant CRT (total radiation dose of 50.4 Gy in 28 fractions, daily dose of 1.8 Gy, with concomitant chemotherapy during the first and last weeks of radiation therapy [5-fluorouracil at the dose of 225 mg/ m 2 per day, 5 days per week]) followed by surgical resection. Eighteen patients were excluded because they had nonresectable tumor (n = 2), metastatic disease (n = 7), or insufficient MR and/or DW image quality (n = 9). In addition, two patients were excluded because they had mucinous tumors; such tumors are identified as lesions with predominantly high T2- weighted signal intensity and exhibit very high ADC values due to very low cellular density. Five patients included in this study were also included in a previous study (30), in which we evaluated the accuracy of ADC measurements in Table 1 Sequence Parameters Parameter T2-weighted Turbo Spin-Echo Imaging DW Imaging* Repetition time (msec) Echo time (msec) No. of echo trains per section 14 1 Matrix size / Field of view (mm) Receiver bandwidth (Hz/pixel) No.of excitations 2 2 Section thickness (mm) Voxel size Distance factor (%) No. of sections Echo-planar imaging factor 150 Acquisition time 2 min 54 sec to 3 min 29 sec 1 min 40 sec * Performed with b values of 50, 400, and 800 sec /mm 2. assessment of rectal cancer downstaging after therapy. The final study group consisted of 62 patients (mean age, 61.5 years [standard deviation]; 41 men, mean age of 61.4 years ; 21 women, mean age of 61.9 years ). MR Imaging Technique All patients underwent both pretreatment MR imaging for primary tumor staging and a second MR imaging examination 7 9 weeks (median time, 58 days) after completion of CRT for restaging of rectal cancer and response evaluation. MR imaging examinations were performed at 1.5 T (Magnetom, Avanto; Siemens Medical Systems, Erlangen, Germany) by using a phasedarray body coil and spine-array coil to optimize signal-to-noise ratio. Before MR imaging, all patients were subjected to bowel cleansing. A dose of 20 mg of the spasmolytic agent hyoscine butilbromide (Buscopan, Boehringer Ingelheim) was administered intravenously to all patients immediately prior to MR imaging to minimize bowel peristalsis and avoid motion artifacts. The imaging protocol consisted of standard T2- weighted turbo spin-echo sequences in three orthogonal directions and an axial DW imaging (single-shot echo-planar imaging) sequence with diffusion sensitivity values (ie, b values of 50, 400, and 800 sec/mm 2 ); the DW imaging sequence was set and angulated identically to the previous axial T2-weighted turbo spin-echo sequence, perpendicular to the tumor axis. Parallel acquisition imaging (generalized autocalibrating partially parallel acquisition, or GRAPPA) was applied to reduce the acquisition time and improve image quality, with acceleration factor of two. ADC maps were generated automatically and included all three b values in a monoexponential decay model. The sequence parameters are displayed in Table 1. Image Analysis and ROI Positioning Methods Pre- and posttreatment T2-weighted MR image sets were analyzed to define the tumor (identified by higher signal intensity compared with the signal intensity of the muscular layer of the adjacent rectal wall); pre- and posttreatment DW images were also analyzed to define the tumor, with tumor being defined as high signal intensity corresponding to the location of the tumor mass on T2-weighted images (Fig 1). Due to the higher resolution of DW images in comparison to ADC maps, ROIs were manually placed on DW images with b of 800 sec/mm 2 first and then copied to the corresponding ADC maps. Post-CRT measurements were 420 radiology.rsna.org n Radiology: Volume 282: Number 2 February 2017

4 Figure 1 Figure 1: Areas of rectal cancer tissue (arrows) on T2-weighted MR images (top row) correspond to highsignal-intensity areas (arrows) on DW images (bottom row) in a 53-year-old male patient before CRT. (Far left images, top and bottom, reprinted from reference 30 under a Creative Commons License.) Figure 2 Figure 2: Pre- and post-crt T2-weighted (left column), pre- and post-crt DW (middle column), and preand post-crt ADC (right column) image sets of rectal cancer in a 53-year-old male patient who experienced complete response to CRT (tumor regression grade 1). Numbers listed on the ADC image are for the particular ROIs shown. (Image at top right, reprinted from reference 30 under a Creative Commons License.) performed with comparison to pre- CRT MR images to ensure that ROIs were placed within the location of the primary tumor. In some patients, highsignal-intensity zones were not identified on post-crt DW images, and then the ROIs were positioned at the location of the tumor bed before CRT (Fig 2). Two radiologists (G.B.L. and I.M.B., with 11 and 6 years of experience in rectal cancer imaging, respectively) independently analyzed pre- and post- CRT MR images, as well as pre- and post-crt DW images, and performed the ADC measurements. The readers were blinded to the clinical patient data, pathology reports, and each other s results. Tumor ADC measurements were obtained by using three different ROI positioning methods: (a) placement of three circular ROIs (the three-rois method); (b) placement of one ROI outlining the tumor on a single section (the single-section method); and (c) placement of an ROI outlining the tumor on each section where it appears (the whole-tumor volume method). All tumor ADC measurements were performed on both pre- and post-crt image sets. For the three-rois method, the mean ADC value was calculated from a sample of three circular ROIs placed within the most cellular tumor areas (ie, areas with the most prominent restrictive diffusion both on DW images and the corresponding ADC maps, with exclusion of T2 shine-through zones) on three different sections containing tumor tissue. The areas of all circular ROIs ranged from 10 mm 2 to 50 mm 2 (median, 22 mm 2 ). For the single-section method, a single freehand ROI was defined by tracing a line around the perceived tumor margins on the section containing the largest tumor area (median ROI area, 881 mm 2 ; range, mm 2 ). For the whole-tumor volume method, freehand ROIs were drawn along the perceived borders of the tumor to cover the entire tumor area on each section containing the tumor (median sum of ROI areas, 4219 mm 2 ; range, mm 2 ), and the mean tumor ADC value was calculated by averaging the measured ADC values for all sections (Fig 3). Histopathologic Analysis The histopathologic tumor regression grade (TRG), as described by Mandard et al (19), served as the standard Radiology: Volume 282: Number 2 February 2017 n radiology.rsna.org 421

5 Figure 3 used to compare the relative diagnostic accuracy between two items with 95% CI, and parameters of diagnostic accuracy (sensitivity, specificity, accuracy, positive predictive value, and negative predictive value) were calculated under the optimal cut-off value for each item, which was determined according to the nearest point to the upper left corner in the ROC curve diagram. Differences in the diagnostic performances of different ROI positioning methods were analyzed by comparing ROC curves according to the method described by DeLong et al (20). All statistical analyses were performed by using Statistical Package for the Social Sciences (IBM SPSS Statistics Software version 20) and MedCalc statistical software (version ). P values less than.05 were considered to indicate a statistically significant difference. Figure 3: Three-ROIs method (top row) and single-section method (bottom row) for ADC measurements in a 62-year-old male patient obtained with circular and freehand ROIs. of reference. TRG 1, or complete regression, referred to rectal tissue specimens without viable tumor cells; TRG 2, to single cells or small groups of cells within rectal tissue specimens; TRG 3, to residual cancer outgrown by fibrosis; TRG 4, to significant fibrosis outgrown by tumor tissue; and TRG 5, to extensive residual cancer without fibrosis. TRG 1 grade was considered complete tumor response to CRT (ypt0), while TRG grades 2 5 were considered noncomplete tumor response to CRT (ypt1 4). Statistical Analysis Interobserver variability for six measurements (performed with three ROI positioning methods before and after CRT) was analyzed by calculating the intraclass correlation coefficient for single measurements with 95% confidence intervals (CIs) (0 0.20, poor correlation; , fair correlation; , moderate correlation; , good correlation; and , excellent correlation). ADC values were averaged between two readers for further analyses. The mean pre- and posttreatment ADC values were compared by using a paired samples t test for each ROI positioning method. The mean ADC values and numeric and percentage values of ADC change obtained by the three different ROI positioning methods were compared by using a nonparametric Wilcoxon rank sum test with regard to the group of patients with complete tumor response (complete response group) and the group of patients with noncomplete tumor response (noncomplete response group). Receiver operating characteristic (ROC) curves were generated for pre- and post-crt measurements as well as for the numeric and percentage values of ADC change for each ROI positioning method to evaluate their diagnostic performance in determining complete tumor response to CRT. The area under the ROC curve (AUC) was considered as relative diagnostic accuracy. Pairwise comparison of the ROC curves was Results Patients and Treatment Characteristics and Histopathologic Findings Forty-six patients underwent a low anterior resection, 13 patients underwent an abdominoperineal resection, and three patients underwent an extended resection. Eighteen tumors were located in the distal rectum, 32 in middle rectum, and 12 in proximal rectum. The median tumor length was 3.2 cm (range, cm), and the median distance from the anal verge was 6.5 cm (range, cm). The median time between the first MR imaging examination and CRT was 26 days (range, days); the median time between the second MR imaging examination and surgery was 10 days (range, 2 23 days). Histopathologic analysis of surgical specimens showed complete tumor response (ypt0, tumor regression grade 1) in 10 specimens and noncomplete tumor response (tumor regression grades 2 5) in 52 specimens. The pretreatment MR-estimated tumor and nodal stages and postoperative histopathologic stages (ypt and ypn) of the study population are summarized in Table radiology.rsna.org n Radiology: Volume 282: Number 2 February 2017

6 Table 2 Pretreatment MR-estimated Tumor and Nodal Stages and Postoperative Histopathologic Stages of the Study Population Stage T0 T1 T2 T3 T4 Total Pretreatment MR imaging and DW-estimated tumor and nodal stages N N N Total Postoperative histopathologic stages (ypt and ypn) N N N Total Effects of Different ROI Positioning Methods on ADC Measurements and Interobserver Variability The mean tumor ADC values acquired with the different ROI positioning methods are displayed in Table 3. By analyzing the whole study population, there were no significant differences between the three-rois and singlesection methods for measurements of numeric and percentage of ADC change (0.26 vs mm 2 / sec, P..99, and 34.0% vs 31.9%, P =.978, respectively) or between the three-rois and whole-tumor volume methods for measurement of percentage of ADC change (34.0% vs 37.3%, P =.249). However, for other measurements and values, significant differences were observed. Interobserver agreement (intraclass correlation coefficient) with the three-rois method was good both for pre-crt and post-crt measurements (0.761 [95% CI: 0.209, 0.930] and [95% CI: 0.164, 0.895], respectively). ICCs were good for pre-crt and moderate for post- CRT measurements made with the single-section method (0.608 [95% CI: 0.287, 0.844] and [95% CI: 0.176, 0.870], respectively); they were excellent for pre-crt and good for post-crt measurements made with the whole-tumor volume method (0.891 [95% CI: 0.615, 0.956] and [95% CI: 0.316, 0.892], respectively). ROI Positioning Methods in the Evaluation of Tumor Response to CRT Table 3 shows the mean tumor ADC values for the pre-crt and post-crt measurements and numeric and percentage values of ADC change between pre- and post-crt measurements for each ROI positioning method in the complete response (CR) and noncomplete response (non-cr) groups. Mean pretreatment ADC values did not differ significantly between the CR and non-cr groups for any of the ROI positioning methods (three-rois method: 0.76 vs mm 2 /sec, P =.278; single-section method: 0.83 vs mm 2 /sec, P =.395; and wholetumor volume method: 0.85 vs mm 2 /sec, P =.181). Post-CRT mean ADC values acquired with the three-rois, single-section, and wholetumor volume methods were significantly higher in the CR group than in the non-cr group (1.13 vs mm 2 /sec, 1.28 vs mm 2 /sec, and 1.36 vs mm 2 /sec, P =.021, P =.002 and P,.001, respectively). Measurements of ADC change, both numeric (DADC) and percentage (%DADC), differed significantly between the CR and non-cr groups regardless of which ROI positioning method was used. Mean numeric change values in the CR and non-cr groups were 0.36 versus mm 2 /sec, P =.013, for the three-rois method; 0.45 versus mm 2 / sec, P =.001, for the single-section method; and 0.51 versus mm 2 /sec, P,.001 for the whole-tumor volume method; mean ADC change percentages were 49.0% versus 31.1%, P =.010, for the three-rois method; 54.5% versus 27.6%, P =.001, for the single-section method; and 61.0% versus 32.7%, P,.001, for the wholetumor volume method. Figure 4 allows comparison of the mean pre- and posttreatment ADC values and numeric and percentage values of ADC change acquired with the different ROI positioning methods in the CR and non-cr groups. Accuracy as reflected by ROC analysis. ROC curves were generated to evaluate diagnostic performances of the different ROI positioning methods in determining complete tumor response to CRT (Fig 5). All pretreatment measurements resulted in AUCs with values of around 0.6 (0.604 [95% CI: 0.423, 0.785] for the three-rois method, [95% CI: 0.417, 0.752] for the single-section method, and [95% CI: 0.454, 0.831] for the whole-tumor volume method), which were not significantly different from each other (P =.930 to..99). The posttreatment measurements resulted in AUCs of 0.731(95% CI: 0.583, 0.878) for the three-rois method, (95% CI: 0.778, 0.970) for the single-section method, and (95% CI: 0.781, 0.990) for the whole-tumor volume method; the AUCs for the single-section and whole-tumor volume methods were similar (P..99) and were both significantly greater than the AUC for the three-rois method (P =.033 and P =.003, respectively). All ROI positioning methods achieved relatively high AUC values in discriminating complete from noncomplete tumor response based on measurements of ADC change, both numeric and percentage (three-rois, single-section, and whole-tumor volume methods: respective AUCs were [95% CI: 0.591, 0.890], [95% CI: 0.812, 0.972], and [95% CI: 0.801, 0.994] for numeric values of ADC change and [95% CI: 0.585, 0.881], [95% CI: 0.798, 0.965], and [95% CI: 0.785, 0.979] for Radiology: Volume 282: Number 2 February 2017 n radiology.rsna.org 423

7 Table 3 Mean Pre- and Post-CRT ADC Values and Mean Numeric and Percentage Values of ADC Change Acquired with Different ROI Positioning Methods ROI Positioning Method percentage ADC change). However, the single-section and whole-tumor volume methods were equally accurate (P =.892 and P..99, for numeric and percentage values, respectively) and significantly better than the three-rois method for numeric ADC change (P =.048 and P =.021, respectively). Also, the AUC for the whole-tumor volume method was significantly greater than for the three-rois method for percentage ADC change (P =.042). All Complete Responders Noncomplete Responders P Value* Three ROIs Pre-CRT ADC ( mm 2 /sec) 1st reader nd reader Average Post-CRT ADC ( mm 2 /sec) 1st reader nd reader Average DADC %DADC Single section Pre-CRT ADC ( mm 2 /sec) 1st reader nd reader Average Post-CRT ADC ( mm 2 /sec) 1st reader nd reader Average ,.001 DADC ,.001 %DADC ,.001 WTV Pre-CRT ADC ( mm 2 /sec) 1st reader nd reader Average Post-CRT ADC ( mm 2 /sec) 1st reader nd reader Average ,.001 DADC ,.001 %DADC ,.001 Note. Data are means 6 standard deviation. WTV = whole-tumor volume. * For differences between the complete response (n = 10) and noncomplete response (n = 52) groups obtained by using a nonparametric Wilcoxon Rank Sum Test. Overall accuracy assessed at optimal cut-off values. In determining complete tumor response based on posttreatment ADC measurements and measurements of ADC change, the single-section and whole-tumor volume methods showed higher accuracy than did the three-rois method; accuracy levels in determining response with posttreatment ADC measurements were 79% (49 of 62) and 77% (48 of 62) versus 61% (38 of 62), with cut-off values of 1.21, 1.30, and mm 2 /sec, respectively; accuracy levels in determining complete response with numeric ADC change were 79% (49 of 62) and 86% (53 of 62) versus 60% (37 of 62), with cut-off values set at increases of 0.33, 0.45, and mm 2 /sec, respectively; accuracy levels in determining complete response with percentage of ADC change were 77% (48 of 62) and 84% (52 of 62) versus 57% (35 of 62), with cut-off values set at increases of 40%, 54%, and 27%, respectively. Table 4 lists accuracy, sensitivity, specificity, and positive and negative predictive values calculated at optimal cut-off levels for all methods. Discussion Assessment of tumor viability on DW images is based on measurements of ADC values within the tumor tissue. The measurements are made from parametric ADC maps, which are generated by mathematical modeling of extracellular movement of water protons in a monoexponential decay model, by using applied diffusion sensitivity values. Quantitative measurements of tumor viability are obtained by defining ROIs on ADC maps; the ROIs can be manually delineated in various sizes in circular or freehand mode by using current commercially available software platforms. In our study, mean rectal tumor ADC values and ADC change values for the whole study population derived with three different ROI positioning methods differed significantly from each other in the majority of comparisons. Thus, in concordance with the results of Lambregts et al, our results indicate that variations in ROI number, size, and position do substantially influence tumor ADC measurements (18). It would therefore be desirable to define an optimal method for measuring ADC values for the assessment of tumor viability. The results of our study show that the three-rois method is inferior to the single-section and whole-tumor volume methods for assessing complete response of rectal cancer to CRT. 424 radiology.rsna.org n Radiology: Volume 282: Number 2 February 2017

8 Figure 4 Figure 4: Box and whisker plots represent distribution of pre- and post-crt ADC values and numeric and percentage values of ADC change between post- and pretreatment measurements obtained with three ROI positioning methods in complete response (CR) and noncomplete response (non CR) groups. Middle line in each box represents the median value of ADC change (pre-crt measurements: 0.76, 0.79, 0.83, 0.85, 0.85, and mm 2 /sec; post-crt measurements: 1.13, 1.03, 1.28, 1.08, 1.32, and mm 2 /sec; numeric value of ADC change [DADC]: 0.36, 0.24, 0.45, 0.23, 0.51, and mm 2 /sec; percentage value of ADC change [%DADC]: 48.9%, 31.1%, 54.5%, 27.6%, 61.0%, and 32.7%). Lower and upper boundaries of the boxes represent the first and third quartiles (25th and 75th percentiles), respectively. Whiskers indicate the maximum and minimum calculated values of ADC change. SS = single section, WTV = whole-tumor volume. Compared with the three-rois method, the use of single-section and whole-tumor volume methods led to significantly higher accuracy in predicting complete response based on post-crt measurements and measurements of numeric ADC change (P = ). Our results indicate that ADC measurements covering all tumor areas, including areas of necrosis and fibrosis, more accurately predict tumor response to therapy than do measurements that include only viable tumor regions. This finding corresponds well with the results of Goh et al (17), who analyzed the influence of different ROI positioning methods on rectal cancer vascular parameter measurements, and of Roth et al (21), who observed that whole-tumor volume measurements better predicted colon carcinoma response to antineoplastic therapies in mice than did ROIs covering only viable tumor areas. Furthermore, our results show that interobserver agreement is dependent on the ROI positioning method used. ADC measurements covering whole-tumor volume showed higher interobserver agreement than those obtained from measurements of smaller tumor areas, and interobserver agreement was excellent for pre-crt measurements acquired with the whole-tumor volume method (intraclass correlation coefficient, [95% CI: 0.615, 0.956]). Also, regardless of the method used, interobserver agreement was higher for pretreatment measurements than for posttreatment measurements, which could be explained by the fact that tumor definition on images is worse after CRT, when tumor has usually undergone fibrotic or necrotic changes or even completely regressed. However, although the point estimates for intraclass correlation coefficient were higher for the whole-tumor volume method and for pre-crt measurements, intraclass correlation coefficient confidence intervals showed large overlap. Our results also indicate that, regardless of the ROI positioning method applied, pretreatment ADC measurements caot reliably predict complete response to CRT; this is consistent with the results of some prior studies (22 24). Some authors reported that pre-crt ADC values were lower for tumors that were responding to therapy than for tumors that were not responding; however, they utilized different criteria to define tumor response. Sun et al (25) defined responders as a T-downstaged group, while Dzik-Jurasz et al (26) used 50% reduction of tumor size as a criterion. With regard to posttreatment measurements, our results align with those of Monguzzi et al (27), who reported similar diagnostic performance in the assessment of response to CRT based on post-crt ADC measurements acquired with a method resembling our single-section method; however, they classified both tumor regression grades 1 and 2 tumors in the responder category. Ha et al (23) measured mean tumor ADC value by placing at least four circular ROIs (each in the range of 16 to 56 mm 2 ) over the tumor area on each section containing the tumor; diagnostic performance with this method of post-crt ADC measurements (AUC, 0.705) was comparable to that found in our study for the three-rois Radiology: Volume 282: Number 2 February 2017 n radiology.rsna.org 425

9 Figure 5 Figure 5: ROC curves show the diagnostic performance of each ROI positioning method in assessment of complete tumor response based on pre-crt (AUC: [95% CI: 0.423, 0.785], [95% CI: 0.417, 0.752], and [95% CI: 0.454, 0.831]; cut-off values: 0.78, 0.86, and mm 2 / sec, respectively), post-crt ADC (AUC: [95% CI: 0.583, 0.878], [95% CI: 0.778, 0.970], and [95% CI: 0.781, 0.990]; cut-off values: 1.12, 1.21, and mm 2 /sec, respectively), numeric ADC change (delta) (AUC: [95% CI: 0.591, 0.890], [95% CI: 0.812, 0.972], and [95% CI: 0.801, 0.994]; cut-off values: +0.28, +0.37, and mm 2 /sec, respectively), and percentage ADC change (% delta) (AUC: [95% CI: 0.585, 0.881], [95% CI: 0.798, 0.965], and [95% CI: 0.785, 0.979]; cut-off values: +31%, +39%, and +54%, respectively). SS = single section, WTV = whole-tumor volume. method (AUC, [95% CI: 0.583, 0.878]). On the other hand, in a recent study by Cai et al, posttreatment DW signal intensity measurements had higher diagnostic accuracy in determining complete tumor response than did posttreatment ADC measurements, which showed unsatisfactory diagnostic performance (AUC, ) (28). These results may be explained by the fact that ADC measurements were obtained by using three circular ROIs that were placed on the highest signal intensity within the tumor tissue, and each ROI covered an area of just 4 mm 2. Our study showed that measurement of larger areas provides more reliable tumor tissue characterization. Furthermore, we specifically avoided measurement of areas with T2 shine-through effect, since inclusion of such areas may result in measurement errors. With regard to determination of complete tumor response based on measurements of ADC change, the results of previous studies are diverse. Kim et al (22) and Sun et al (25) found that the mean percentage of tumor ADC change was a useful predictor of tumor response to therapy. Semedo et al (29) reported insufficient diagnostic performance of ADC measurements in the estimation of complete tumor response to CRT, with an AUC of 0.51 for the percentage of ADC change. Monguzzi et al (27) found that the diagnostic performance of the percentage of ADC change was inferior to that of numeric ADC change (AUC, vs , respectively). To our knowledge, ours is the only study to date in which measurements were acquired in three different ways to predict complete tumor response to CRT, and the percentage of ADC change measured with the whole-tumor volume method achieved the highest diagnostic performance, with an AUC of (95% CI: 0.785, 0.979), sensitivity of 80% (eight of 10; 95% CI: 44.4%, 97.5%), specificity of 85% (44 of 52; 95% CI: 71.9%, 93.1%), accuracy of 84% (52 of 62), positive predictive value of 50% (eight of 16), and negative predictive change of 96% (44 of 46), by using a 54% ADC increase as the cut-off value. We believe several of our results are clinically important. First, with all three positioning methods, both post-crt ADC values and ADC change values performed well in discriminating noncomplete responders eligible for surgery; this was manifested by very high negative predictive values ranging from 96% (44 of 46) to 100% (39 of 39). Furthermore, diagnostic performance levels in the assessment of complete tumor response to therapy were comparable with the single-section and whole-tumor volume methods and for post-crt measurements and ADC change measurements; this suggests that obtaining post-crt measurements with the singlesection method, which can be done relatively quickly, may be the most sensible approach for assessing response in clinical practice (for this approach, AUC = [95% CI: 0.778, 0.970], accuracy = 79% [49 of 62], sensitivity = 90% [nine of 10; 95% CI: 55.5%, 99.7%], specificity = 77% [40 of 52; 95% CI: 63.2%, 87.5%]). Also noteworthy was the high diagnostic performance of the whole-tumor volume method for prediction of complete response based on the percentage of ADC change (detailed 426 radiology.rsna.org n Radiology: Volume 282: Number 2 February 2017

10 Table 4 Parameters of Diagnostic Performance of All ROI Positioning Methods for Determining Complete Tumor Response Based on Different ADC Measurements Method Sensitivity (%) Specificity (%) PPV (%) NPV (%) Accuracy (%) AUC* Optimal Cut-off Value Pre-CRT ADC ( mm 2 /sec) Three ROIs 70 (7/10) 54 (28/52) 23 (7/31) 90 (28/31) 57 (35/62) (0.423, 0.785) 0.78 Single section 80 (8/10) 52 (27/52) 24 (8/33) 93 (27/29) 57 (35/62) (0.417, 0.752) 0.86 WTV 70 (7/10) 69 (36/52) 30 (7/23) 92 (36/39) 69 (43/62) (0.454, 0.831) 0.85 Post-CRT ADC ( mm 2 /sec) Three ROIs 90 (9/10) 55 (29/52) 28 (9/32) 97 (29/30) 61 (38/62) (0.583, 0.878) 1.05 Single section 90 (9/10) 77 (40/52) 43 (9/21) 98 (40/41) 79 (49/62) (0.778, 0.970) 1.21 WTV 90 (9/10) 75 (39/52) 41 (9/22) 98 (39/40) 77 (48/62) (0.781, 0.990) 1.30 DADC ( mm 2 /sec) Three ROIs 90 (9/10) 54 (28/52) 27 (9/33) 97 (28/29) 60 (37/62) (0.591, 0.890) Single section 100 (10/10) 75 (39/52) 44 (10/23) 100 (39/39) 79 (49/62) (0.812, 0.972) WTV 90 (9/10) 85 (44/52) 53 (9/17) 98 (44/45) 86 (53/62) (0.801, 0.994) %DADC Three ROIs 100 (10/10) 48 (25/52) 27 (10/37) 100 (25/25) 57 (35/62) (0.585, 0.881) +27 Single section 100 (10/10) 73 (38/52) 41 (10/24) 100 (38/38) 77 (48/62) (0.798, 0.965) +40 WTV 80 (8/10) 85 (44/52) 50 (8/16) 96 (44/46) 84 (52/62) (0.785, 0.979) +54 Note. ROC curves were generated for all ADC measurements for each ROI positioning method. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy were calculated under the optimal cut-off value listed for each item, which was determined according to the nearest point to the upper left corner in the ROC curve diagram. WTV = whole-tumor volume. * Data in parentheses are 95% CIs. earlier). However, none of the methods yielded a PPV higher than 53% (nine of 17), indicating that none is completely reliable for the selection of complete responders eligible for less invasive treatment, probably due to the considerable overlap of ADC values between the complete response and noncomplete response groups. Our study had some limitations. ADC measurements are very subtle and subject to measurement error. In some patients, little or no residual tumor could be identified on DW images after CRT, so it was challenging to place the ROI(s) properly. Furthermore, we excluded nine patients whose DW images were highly distorted by susceptibility artifacts from intraluminal air, as these artifacts would have affected measurements. Another limitation was that ADC maps were acquired with a monoexponential algorithm; a multiple exponential fit with more b values may have been more accurate and preferable. Also, the cut-off values we evaluated in the current study were not prespecified but instead were based on data from this test set an approach that is likely to result in overestimation of the diagnostic performance of the tested methods. The thresholds derived from this study need to be evaluated in a separate, prospective validation study. Finally, the number of patients with pathologic complete tumor response was small and comprised just 16% (10 of 62) of the study population. Our study has shown that the ROI positioning method used significantly affects ADC measurements. Measurements of ADC change induced by CRT may have considerable diagnostic value for the estimation of complete tumor response, and the highest accuracy is obtained by analyzing the whole-tumor volume and the percentage of ADC change. With each of the three ROI positioning methods, posttreatment ADC measurements and measurements of ADC change demonstrated excellent performance in distinguishing noncomplete responders who require surgical treatment. Disclosures of Conflicts of Interest: I.M.B. disclosed no relevant relationships. G.B.L. disclosed no relevant relationships. M.M.G. disclosed no relevant relationships. References 1. O Neill BD, Brown G, Heald RJ, Cuingham D, Tait DM. Non-operative treatment after neoadjuvant chemoradiotherapy for rectal cancer. Lancet Oncol 2007;8(7): Habr-Gama A, Perez RO, Proscurshim I, et al. Patterns of failure and survival for nonoperative treatment of stage c0 distal rectal cancer following neoadjuvant chemoradiation therapy. J Gastrointest Surg 2006;10(10): ; discussion Habr-Gama A, Perez RO, Wy G, Marks J, Kessler H, Gama-Rodrigues J. Complete clinical response after neoadjuvant chemoradiation therapy for distal rectal cancer: characterization of clinical and endoscopic findings for standardization. Dis Colon Rectum 2010;53(12): Maas M, Nelemans PJ, Valentini V, et al. Long-term outcome in patients with a pathological complete response after chemoradiation for rectal cancer: a pooled analysis of individual patient data. Lancet Oncol 2010; 11(9): Maas M, Beets-Tan RG, Lambregts DM, et al. Wait-and-see policy for clinical complete responders after chemoradiation for rectal cancer. J Clin Oncol 2011;29(35): Beets-Tan RG, Lambregts DM, Maas M, et al. Magnetic resonance imaging for the clinical Radiology: Volume 282: Number 2 February 2017 n radiology.rsna.org 427

11 management of rectal cancer patients: recommendations from the 2012 European Society of Gastrointestinal and Abdominal Radiology (ESGAR) consensus meeting. Eur Radiol 2013;23(9): Gollub MJ, Gultekin DH, Akin O, et al. Dynamic contrast enhanced-mri for the detection of pathological complete response to neoadjuvant chemotherapy for locally advanced rectal cancer. Eur Radiol 2012;22(4): Goh V, Padhani AR, Rasheed S. Functional imaging of colorectal cancer angiogenesis. Lancet Oncol 2007;8(3): Song I, Kim SH, Lee SJ, Choi JY, Kim MJ, Rhim H. Value of diffusion-weighted imaging in the detection of viable tumour after neoadjuvant chemoradiation therapy in patients with locally advanced rectal cancer: comparison with T2 weighted and PET/CT imaging. Br J Radiol 2012;85(1013): Lambrecht M, Deroose C, Roels S, et al. The use of FDG-PET/CT and diffusionweighted magnetic resonance imaging for response prediction before, during and after preoperative chemoradiotherapy for rectal cancer. Acta Oncol 2010;49(7): Lambregts DM, Vandecaveye V, Barbaro B, et al. Diffusion-weighted MRI for selection of complete responders after chemoradiation for locally advanced rectal cancer: a multicenter study. A Surg Oncol 2011;18(8): Charles-Edwards EM, desouza NM. Diffusion-weighted magnetic resonance imaging and its application to cancer. Cancer Imaging 2006;6: Padhani AR, Liu G, Koh DM, et al. Diffusionweighted magnetic resonance imaging as a cancer biomarker: consensus and recommendations. Neoplasia 2009;11(2): Sun YS, Cui Y, Tang L, et al. Early evaluation of cancer response by a new functional biomarker: apparent diffusion coefficient. AJR Am J Roentgenol 2011;197(1):W23 W De Cobelli F, Giganti F, Orsenigo E, et al. Apparent diffusion coefficient modifications in assessing gastro-oesophageal cancer response to neoadjuvant treatment: comparison with tumour regression grade at histology. Eur Radiol 2013;23(8): Kim YC, Lim JS, Keum KC, et al. Comparison of diffusion-weighted MRI and MR volumetry in the evaluation of early treatment outcomes after preoperative chemoradiotherapy for locally advanced rectal cancer. J Magn Reson Imaging 2011;34(3): Goh V, Halligan S, Gharpuray A, Wellsted D, Sundin J, Bartram CI. Quantitative assessment of colorectal cancer tumor vascular parameters by using perfusion CT: influence of tumor region of interest. Radiology 2008;247(3): Lambregts DM, Beets GL, Maas M, et al. Tumour ADC measurements in rectal cancer: effect of ROI methods on ADC values and interobserver variability. Eur Radiol 2011; 21(12): Mandard AM, Dalibard F, Mandard JC, et al. Pathologic assessment of tumor regression after preoperative chemoradiotherapy of esophageal carcinoma: clinicopathologic correlations. Cancer 1994;73(11): DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 1988;44(3): Roth Y, Tichler T, Kostenich G, et al. Highb-value diffusion-weighted MR imaging for pretreatment prediction and early monitoring of tumor response to therapy in mice. Radiology 2004;232(3): Kim SH, Lee JY, Lee JM, Han JK, Choi BI. Apparent diffusion coefficient for evaluating tumour response to neoadjuvant chemoradiation therapy for locally advanced rectal cancer. Eur Radiol 2011;21(5): Ha HI, Kim AY, Yu CS, Park SH, Ha HK. Locally advanced rectal cancer: diffusionweighted MR tumour volumetry and the apparent diffusion coefficient for evaluating complete remission after preoperative chemoradiation therapy. Eur Radiol 2013; 23(12): DeVries AF, Kremser C, Hein PA, et al. Tumor microcirculation and diffusion predict therapy outcome for primary rectal carcinoma. Int J Radiat Oncol Biol Phys 2003; 56(4): Sun YS, Zhang XP, Tang L, et al. Locally advanced rectal carcinoma treated with preoperative chemotherapy and radiation therapy: preliminary analysis of diffusionweighted MR imaging for early detection of tumor histopathologic downstaging. Radiology 2010;254(1): Dzik-Jurasz A, Domenig C, George M, et al. Diffusion MRI for prediction of response of rectal cancer to chemoradiation. Lancet 2002; 360(9329): Monguzzi L, Ippolito D, Bernasconi DP, Trattenero C, Galimberti S, Sironi S. Locally advanced rectal cancer: value of ADC mapping in prediction of tumor response to radiochemotherapy. Eur J Radiol 2013; 82(2): Cai PQ, Wu YP, An X, et al. Simple measurements on diffusion-weighted MR imaging for assessment of complete response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer. Eur Radiol 2014;24(11): Curvo-Semedo L, Lambregts DM, Maas M, et al. Rectal cancer: assessment of complete response to preoperative combined radiation therapy with chemotherapy--conventional MR volumetry versus diffusion-weighted MR imaging. Radiology 2011;260(3): Blazic I, Maksimovic R, Gajic M, Saranovic D. Apparent diffusion coefficient measurement covering complete tumor area better predicts rectal cancer response to neoadjuvant chemoradiotherapy. Croatian Med J 2015;56(5): radiology.rsna.org n Radiology: Volume 282: Number 2 February 2017