Can the preoperative CT-based deep learning radiomics model predict histologic grade and prognosis of chondrosarcoma?

Computed tomography (CT) and biopsy may be insufficient for preoperative evaluation of the grade and outcome of patients with chondrosarcoma. The aim of this study was to develop and validate a CT-based deep learning radiomics model (DLRM) for predicting histologic grade and prognosis in chondrosarcoma (CS).

A multicenter 211 (training cohort/ test cohort, 127/84) CS patients were enrolled. Radiomics signature (RS), deep learning signature (DLS), and DLRM incorporating radiomics and deep learning features were developed for predicting the grade. Kaplan-Meier survival analysis was used to assess the association of the model-predicted grade with recurrence-free survival (RFS). Model performance was evaluated with the area under the receiver operating characteristic curve (AUC) and the Harrell's concordance index (C-index).

The DLRM (AUC, 0.879; 95 % confidence interval [CI], 0.802-0.956) outperformed (z = 2.773, P=0.006) the RS (AUC, 0.715;95 % CI, 0.606-0.825) in predicting grade in the test cohort. RFS showed significant differences (log-rank test, P<0.05) between low-grade and high-grade patients stratified by DLRM. The DLRM achieved a higher C-index (0.805; 95 % CI, 0.694-0.916) than the RS (0.692, 95 % CI, 0.540-0.844) did in predicting RFS for CS patients in the test cohort.

The DLRM can accurately predict the histologic grade and prognosis in CS.