Deep learning for diagnosis and survival prediction in soft tissue sarcoma

•A DLM was able to classify five of the most common STS subtypes from histology alone.•When aided by DLM, pathologists were more accurate, faster, and certain in their diagnosis.•A similar predict disease-specific survival status subtype.•The DLM's prediction an independent prognostic factor.•New image features associated with could be identified. BackgroundClinical management soft tissue sarcoma (STS) is particularly challenging. Here, we used digital pathology deep learning (DL) for diagnosis prognosis STS.Patients methodsOur retrospective, multicenter study included a total 506 histopathological slides 291 patients STS. The Cancer Genome Atlas cohort (240 patients) served as training validation set. A second, (51 additional test use DL model (DLM) clinical decision support system evaluated nine different levels expertise. For prediction, 139 85 leiomyosarcoma (LMS) used. Area under receiver operating characteristic (AUROC) accuracy main outcome measures.ResultsThe achieved mean AUROC 0.97 (±0.01) 79.9% (±6.1%) diagnosing subtypes. significantly improved 46.3% (±15.5%) 87.1% (±11.1%). Furthermore, they faster diagnosis. In LMS, predicting 0.91 (±0.1) 88.9% (±9.9%). Cox regression showed DLM’s significant factor (P = 0.008, hazard ratio 5.5, 95% confidence interval 1.56-19.7) these patients, outperforming other risk factors.ConclusionsDL can accurately diagnose frequent conventional slides. It might prevalent subtype our cohort. also help make accurate diagnoses. This substantially improve patients. Clinical Our measures. factors.