guiqin Liu¹, shuaishuai XU¹, yongming Dai², Guangyu WU¹, and jianrong XU^1
1Radiology, Renji Hospital，Shanghai Jiaotong University School of Medicine, Shanghai, China, ²United Imaing Healthcare, Shanghai, China

Purpose: To investigate the value of radiomics features from diffusion-weighted imaging (DWI) in differentiating muscle-invasive bladder cancer (MIBC) from non-muscle-invasive bladder cancer (NMIBC). Methods: This retrospective study included 218 pathologically-confirmed bladder cancer patients (training set: 131 patients, 86 MIBC; validation set: 87 patients, 55 MIBC) who underwent DWI before biopsy through transurethral resection (TUR) between July 2014 and December 2018. Radiomics models based on DWI for discriminating state of muscle-invasive were built using random forest (RF) and all-relevant (AR) methods on the training set and were tested on validation set. Combination models based on TUR data were also built. Discrimination performances were evaluated with the area under the receiver operating characteristic (ROC) curve (AUC), accuracy, sensitivity, specificity, F1 and F2 scores. Qualitative MRI evaluation based on morphology was performed for comparison. Results: No significant difference was found between RF and AR models. RF model was more sensitive than TUR (0.873 vs 0.655, p=0.019) for discriminating muscle-invasive bladder cancer. When combining RF with TUR, the sensitivity increased to 0.964, significantly higher than TUR (0.655, p<0.001), MRI evaluation (0.764, p=0.006), and the combination of TUR and MRI (0.836, p=0.046). Combining RF and TUR achieved the highest accuracy of 0.897 and F2 score of 0.946. Conclusion: Combining DWI radiomics features with TUR could improve the sensitivity and accuracy in discriminating the presence of muscle invasion in bladder cancer for clinical practice. Multi-center, prospective studies are needed to confirm our results.