Our purpose was to determine the value of %△V and %△SIR from multiparametric MRI, using individual sequences and sequence combinations, for predicting preoperative TRG after CRT in the patients with LARC. Our results indicated good to excellent diagnostic performance from T2W-%△V, DW-%△V, ceT1W-%△V, T2W-%△SIR, ceT1W-%△SIR, and DW-%△V * T2W-%△SIR for discriminating TRG0 from TRG1, TRG0 from TRG2, and TRG0 from TRG3; and for discriminating TRG1 from TRG3. Although our ROC analysis indicated no significant differences among these AUC values, DW-%△V * T2W-%△SIR had highest AUC values. In addition, none of the measurements were useful for distinguishing TRG1 from TRG2, nor for distinguishing TRG2 from TRG3. To the best of our knowledge, no prior studies comprehensively compared the diagnostic performances of change in V and SIR using T2W, DWI, and ceT1W MRI sequences to predict TRG in patients with LARC after CRT. Predicting preoperative TRG may help clinicians to stratify patients by prognosis and to implement optimal preoperative therapy(4).
We found a low agreement of post-CRT MR T restaging and ypT stage (κ = 0.191), consistent with the results of some prior studies that mrTRG cannot be used as a surrogate to predict pTRG(3, 7, 11). Previous research reported that MR quantitative results of tumor V and SI were important prognostic indicators for tumor change during CRT and response after CRT. In particular, van den Begen et al. and Lambregts et al. reported that visual changes in rectal tumor morphology (fibrosis) and volume occurred during CRT(14, 19). Tumor V correlated with downstaging of rectal cancer(17, 30), and Lambregts et al. reported that post-CRT DWI volumetry provided the best results for detection of patients with a CR after CRT (AUC = 0.92, sensitivity = 70%, specificity = 98%)(17). In line with previous studies, we found that tumor V had excellent diagnostic performance in discriminating TRG0 from TRG1, TRG0 from TRG2, and TRG0 from TRG3 (AUC = 0.921 ~ 0.984, sensitivity = 75 ~ 93.75%, and specificity = 85.71 ~ 100%).
Hötker et al. determined tumour volumetry from post-treatment dynamic contrast enhanced (DCE)-MRI and DW-MRI, and reported good correlations with histopathological percent tumour regression in resected specimens, and performance that was superior to post-CRT T2 tumour volumetry(11). In the present MRI study, the diagnostic performance of tumor V from DW was slightly greater than that from T2W and ceT1W, although there were no significant differences among AUC values. The better performance of DW images may be because this method provides easier recognition of viable tumor remnants, because they appear hyperintense compared with the low SI of surrounding non-neoplastic tissue. These results are in contrast with previous studies which showed that post-CRT DW MR volumetry provided excellent diagnostic performance in assessing CR (AUC = 0.93) and was significantly more accurate than T2-weighted MR volumetry(16).
CRT may lead to no change or only a slight change in tumour V in poor responders, but there may be fibrotic transformation that cannot be visually identified(31). The measured value of SIR may have different changes, because SIR and V can respond differently to CRT(22–25). Wan et al reported that T2WI SI related parameters, which had AUC ranging from 0.694 to 0.762, sensitivity from 68.2 to 77.3%, and specificity from 63.6 to 77.0%, were potential predictors for pCR in LARC patients after CRT(15). The value of intra-tumor heterogeneity evaluated by DW for predicting TRG after CRT in the lower rectum and measurements of ADC change induced by CRT may have considerable diagnostic value for the estimation of CR(25, 32). DCE-MRI in rectal cancer is promising, mainly for prediction of prognosis and assessment of response to CRT(33). Our results are in contrast with previous studies, in that we identified %ΔSIR on T2W and ceT1W as promising diagnostic tools for distinguishing CR from non-CR and TRG1 from TRG3, respectively. However, our ADC-%△SIR results were useless, and this discrepancy from published data is probably due to the use of different protocols, different methods for ROI selection, and differences in factors contributing to magnetic field inhomogeneity, such as pH, hydration status, and susceptibility effects(25, 32, 34, 35).
Previous studies used SIR and V from MRI sequences to discriminate pCR or response groups from non-pCR or nonresponsive groups, not for discrimination of different TRG groups(2, 3, 6, 11, 15, 19, 21, 25, 32, 35). In contrast, we used changes of SIR and V from T2W, DW/ADC, and ceT1W sequences to differentiate different TRGs after CRT. Our results indicated that T2W-%△V, DW-%△V, ceT1W-%△V, T2W-%△SIR, ceT1W-%△SIR, and DW-%△V * T2W-%△SIR provided high diagnostic performances, with accuracies of 82.35 to 100%, in prediction of TRG after pCRT, better than reported in previous studies(2, 6, 14–19, 25, 27, 30, 32, 33). In discriminating TRG0 from TRG2, the AUC for ceT1W-%△SIR was significantly lower than the AUC for other sequences (p < 0.05). In discriminating pCR from non-pCR and TRG1 from TRG3, there were no significantly different AUC values from using different parameters. This indicates that multiple parameters provided high diagnostic performance for discriminating pCR from non-pCR and TRG1 from TRG3. DW-%△V * T2W-%△SIR had the best prediction of TRG, with AUCs of 0.954 ~ 1.000, sensitivity of 93.75%~100%, specificity of 97.14 ~ 100%, and accuracy of 96.08%~100%. %△V and %△SIR did not distinguished TRG1 from TRG 2, nor TRG2 from TRG3, possibly because there were only small differences in V and fibrotic transformation between these TRGs or because there were changes in ROI selection after CRT. The V parameter can distinguish TRG0 from TRG1, possibly because the ROI area is smaller in TRG0 than TRG1; the ROI for TRG0 has a fibrous bed but no tumor, but the ROI of TRG1 contains some tumor and some fibrous bed.
Although our results are encouraging, there were some limitations to our study. First, this was a retrospective study and there were small numbers of patients in the different TRG groups, and this could have led to selection bias. Thus, it is necessary to examine more patients and use a prospective study to confirm our results. Second, our data were from multiple MRI machines because it was collected retrospectively. However, except for the fair inter-observer correlation of TRG with ADC-%△SIR, other MRI measurements that had excellent inter-observer correlation(2, 17, 25). Previous studies reported better reproducibility of SIR than SI for several MRI sequences(2, 16, 18, 22–24, 34, 35). Moreover, Hötker et al. reported that DCE-MRI volumetry had better inter-observer agreement(11). Third, the selection of an ROI and comparison of MRI results before and after CRT was somewhat subjective; however, to overcome these limitations, we evaluated a relatively large area of cancerous tissue(32). Blazic et al. reported that the use of single-section and whole-tumor volume methods provided similar accuracy in predicting CR based on post-CRT measurement of ADC change, and that use of single sections was less time-consuming(25). Finally, there were variations in the timing between the pre- and post-CRT MRI examinations, with a large standard deviation, and no uniform and standard CRT scheme. These factors could have increased the risk of selection bias. However, we used the rate of change in V and SIR, and the degree of tumor change after CRT was incorporated into the corresponding TRG groups, and did not influence grouping of cases by TRG(15, 16).
In conclusion, despite the several limitations of this study, T2W-%△V, DW-%△V, ceT1W-%△V, T2W-%△SIR, ceT1W-%△SIR, and especially DW-%△V* T2W-%△SIR, provided good to excellent diagnostic performance regarding TRG prediction after CRT in patients with LARC. These results require confirmation by a large prospective cohort study. The advantages of establishing the preoperative TRG in these patients are that it provides more effective management that can be specifically tailored to individual patients, it provides reliable predictions of prognosis after CRT, and it may indicate the suitability of non-operable management, such as further CRT.