Journal: Scientific reports
This study presents SarcNet, a multimodal deep-learning algorithm that combines histological whole-slide images and clinical data to predict metastatic relapse in limb and trunk wall soft tissue sarcoma (STS) patients.
Key points about SarcNet include:
- Development: Built using federated learning across two independent cohorts (total n=611).
- Validation: Tested on two additional cohorts (n=217).
- Performance: Achieved a 5-year metastasis-free survival prediction AUC of 0.797 in cross-validation.
- Comparison: Performance is comparable to the Sarculator nomogram and superior to the FNCLCC grading system.
- Interpretability: Identified key histological features influencing predictions, including atypia, tumor cellularity, and anisokaryosis.
SarcNet shows promise as a clinical tool to stratify high-risk patients who may benefit from neoadjuvant or adjuvant therapies.