Construction and clinical validation of a cascaded deep learning system for classification of benign and malignant solid small renal masses based on MRI.
Authors
Affiliations (10)
Affiliations (10)
- Department of Radiology, the First Medical Center, Chinese PLA General Hospital, No.28 Fuxing Road, Haidian District, Beijing 100853, China.
- School of Medical and Information Engineering, Gannan Medical University, No.1, Harmony Avenue, Rongjiang New District, Ganzhou, Jiangxi 341000, China.
- Department of Radiology, the Seventh Medical Center, Chinese PLA General Hospital, No.5 Nanmen Cang, Dongcheng District, Beijing 100853, China.
- Department of CT and MRl, Tonghua Central Hospital, Tonghua, Jilin 134000, China.
- Radiology Department, Peking University First Hospital, No.8 Xishiku Road, Xicheng District, Beijing 100034, China.
- Department of Radiology, Beijing Friendship Hospital, Capital Medical University, No.95 Yong'an Road, Xicheng District, Beijing 100050, China.
- Department of Pathology, the First Medical Center, Chinese PLA General Hospital, No.28 Fuxing Road, Haidian District, Beijing 100853, China.
- Beijing Key Laboratory of Brain Imaging and Connectomics, Beijing Normal University, No.28 Fuxing Road, Haidian District, Beijing 100875, China.
- Hospital Management Institute, Department of Innovative Medical Research, Chinese PLA General Hospital, Outpatient Building, No.28 Fuxing Road, Haidian District, Beijing 100853, China.
- Department of Radiology, the First Medical Center, Chinese PLA General Hospital, No.28 Fuxing Road, Haidian District, Beijing 100853, China. Electronic address: [email protected].
Abstract
This study aims to develop a cascaded deep learning (DL) system based on multiparametric MRI to establish an automated pipeline for the segmentation and classification of small renal masses (SRMs). A retrospective collection of SRM patients with pathologically confirmed from three institutions was conducted. MRI data from Institution 1 were randomly divided into a training set and an internal test set. Data from other institutions served as the external test set. A cascaded DL system was developed, incorporating automated segmentation and benign-malignant classification. Diagnostic performance was evaluated using receiver operating characteristic analysis and compared against three radiologists of varying experience. A total of 965 patients with SRM were included. Institution 1 contributed 888 cases, with 712 used for training and 176 as an internal test set; Institutions 2 and 3 provided 77 cases as an external test set. The optimal classification model using automated segmentation labels achieved AUCs of 0.936 and 0.788 on internal and external test sets, respectively. Performance was comparable to models using manual segmentation (internal: 0.936 vs. 0.944, P = 0.671; external: 0.788 vs. 0.832, P = 0.629). On the external test set, the model performed comparably to the senior radiologist, while it significantly outperformed the senior radiologist on the internal test set. The model significantly outperformed the junior radiologist on both test sets. This finding remained consistent in the subgroup of tumors smaller than 3 cm. The cascaded DL system demonstrated robust performance across multiple centers, enabling non-invasive and efficient discrimination of SRM malignancy, showing promise as a clinical support tool.