Predicting Response to Neoadjuvant Chemotherapy in Ovarian Cancer from CT Baseline Using Multi-Loss Deep Learning

2026-05-14

AI Summary

This study addresses the critical need for early identification of ovarian cancer patients who are unlikely to respond to neoadjuvant chemotherapy, thereby enabling timely transition to primary surgery and avoiding ineffective treatment. Leveraging pre-treatment contrast-enhanced CT scans, the authors employ an automated 3D lesion segmentation mask to extract axial slice features using a pretrained image encoder, which are then aggregated into a volumetric embedding via an attention mechanism. A novel multi-loss strategy—integrating classification loss, supervised contrastive regularization, and hard negative mining—is introduced to enhance discrimination, particularly for ambiguous responders. The proposed model achieves a test-set ROC-AUC of 0.73 (95% CI: 0.58–0.86) and an F1 score of 0.70 (95% CI: 0.56–0.82), demonstrating improved performance in predicting non-response.

Abstract

Ovarian cancer is the most lethal gynecologic malignancy: around 60% of patients are diagnosed at an advanced stage, with an associated 5-year survival rate of about 30%. Early identification of non-responders to neoadjuvant chemotherapy remains a key unmet need, as it could prevent ineffective therapy and avoid delays in optimal surgical management. This work proposes a non-invasive deep learning framework to predict neoadjuvant chemotherapy response from pre-treatment contrast-enhanced CT by leveraging automatically derived 3D lesion masks. The approach encodes axial slices with a partially fine-tuned pretrained image encoder and aggregates slice-level representations into a volumetric embedding through an attention-based module. Training combines classification loss with supervised contrastive regularization and hard-negative mining to improve separation between ambiguous responders and non-responders. The method was developed on a retrospective single-center cohort from the European Institute of Oncology (Milan, IT), including 280 eligible patients (147 responder, 133 non-responder). On the test cohort, the model achieved a ROC-AUC of 0.73 (95% CI: 0.58-0.86) and an F1-score of 0.70 (95% CI: 0.56-0.82). Overall, these results suggest that the proposed architecture learns clinically relevant predictive patterns and provides a robust foundation for an imaging-based stratification tool.

Problem

• ovarian cancer
• neoadjuvant chemotherapy
• treatment response prediction
• non-responders
• CT imaging

Innovation

• multi-loss deep learning
• attention-based aggregation
• supervised contrastive regularization
• hard-negative mining
• 3D lesion mask