Construction of prognostic scoring model for ovarian cancer based on deep learning algorithm

Area of Science:

• Oncology
• Digital Pathology
• Biomedical Imaging

Background:

• Ovarian cancer is a leading cause of cancer death in women, with poor outcomes often linked to late-stage diagnosis.
• Accurate prognostic prediction is crucial for effective ovarian cancer management and treatment planning.

Purpose of the Study:

• To develop and validate a prognostic prediction model for ovarian cancer utilizing histopathological images.
• To assess the model's performance in predicting patient prognosis and its potential for clinical application.

Main Methods:

• Pathological slides from 158 in-house and 105 TCGA-OV cases were processed using Macenko's stain normalization and patch extraction.
• The CLAM framework was employed to construct the prognostic model, validated with time-dependent ROC and survival analyses.
• The model's predictive capability was further evaluated when integrated with clinical and transcriptomic data.

Main Results:

• The prognostic model achieved high predictive accuracy with AUCs of 0.93 (internal validation) and 0.70 (external validation).
• The integrated analysis revealed significant prognostic differences between high- and low-risk patient groups.
• The model demonstrated potential for clinical translation in ovarian cancer prognosis.

Conclusions:

• The developed image-based prognostic model accurately predicts ovarian cancer patient outcomes.
• This tool offers valuable reference for clinical diagnosis and treatment strategies.
• Integration with biomarkers like CA-125 enhances personalized risk stratification for ovarian cancer patients.