Deep Learning-based Fully-Automated Detection and Segmentation of Small Renal Masses on Multi-sequences MRI: A Multi-center Study

Motivation: Automated detection and segmentation method could serve as a fundamental step for diagnosis of small renal mass (SRM)Goal(s): To develop and assess automated segmentation method for SRM using a deep learning method based on multi-sequences MRIApproach: A total of 913 SRM patients from three institutions was used in deep learning model training and testing for five sequences (T2WI, T1WI, CP, NP, DP). The model was evaluated on internal and external test set using DSC (dice similarity coefficient)Results: The overall median DSC of five sequences (T2WI, T1WI, CP, NP, and DP) yield 0.824, 0.769, 0.845, 0.847, 0.855 on whole test set. Impact: The value of radiomics in preoperative diagnosis of benign and malignant SRM had been proven. However, manual segmentation impeded the conduction of radiomics. Automated segmentation models could help efficiently build radiomics model and reduce radiologists’ workloads.