Tumor cell- and infiltrating immune cell-based supervised learning artificial intelligence multimodal platform for...

Area of Science:

• Digital pathology
• Oncology
• Artificial Intelligence

Background:

• Accurate survival prediction for oral squamous cell carcinoma (OSCC) is a significant clinical challenge.
• Existing prognostic tools require enhancement for personalized patient management.

Purpose of the Study:

• To develop and evaluate novel artificial intelligence (AI) platforms for assessing overall survival in OSCC patients.
• To compare AI-based survival prediction with conventional clinical signatures.

Main Methods:

• Utilized 240 whole-slide images from multicenter cohorts for AI model development.
• Implemented supervised deep learning (DL) with annotations (PathS model) and weakly supervised DL using slide-level labels.
• Employed gradient-weighted class activation mapping for feature identification.

Main Results:

• Supervised DL (PathS model) achieved a c-index of 0.809, outperforming weakly supervised DL (c-index=0.707) and conventional clinical signatures (CS model, c-index=0.721).
• AI identified tumor cells and tumor-infiltrating immune cells as key prognostic features.
• A multimodal nomogram combining PathS signatures with CS improved survival prediction accuracy (c-index=0.817).

Conclusions:

• Novel AI platforms demonstrate significant potential for improving overall survival assessment in OSCC.
• The multimodal nomogram offers a substantial advancement in personalized survival prediction for OSCC patients.
• AI-driven insights into prognostic features enhance understanding of OSCC progression.