Predicting endometrial cancer subtypes and molecular features from histopathology images using multi-resolution deep...

Area of Science:

• Pathology
• Artificial Intelligence
• Genomics

Background:

• Accurate classification of endometrial carcinoma into histological and molecular subtypes, along with mutation status, is crucial for patient prognosis and treatment selection.
• While sequencing provides detailed molecular information, it is time-consuming and expensive.
• Current diagnostic methods may not fully capture the molecular complexity essential for personalized medicine.

Purpose of the Study:

• To develop and validate a deep convolutional neural network (CNN) capable of predicting endometrial carcinoma histological subtypes, molecular subtypes, and common gene mutations from H&E-stained images.
• To assess the accuracy and generalizability of the proposed AI model on independent datasets.
• To explore the potential of AI-driven image analysis to complement or replace traditional sequencing methods for endometrial carcinoma subtyping and mutation detection.

Main Methods:

• Implementation of a customized multi-resolution deep convolutional neural network named Panoptes.
• Training and testing the Panoptes model on digitized H&E-stained pathological images of endometrial carcinoma.
• Evaluation of the model's performance in predicting histological subtypes, molecular subtypes, and 18 common gene mutations.

Main Results:

• The Panoptes model demonstrated high accuracy in predicting both histological and molecular subtypes of endometrial carcinoma.
• The AI tool successfully identified 18 common gene mutations directly from pathological images.
• The model exhibited strong generalization capabilities when tested on independent datasets, indicating robustness.

Conclusions:

• The developed AI model, Panoptes, shows significant potential for clinical application in endometrial carcinoma diagnostics.
• Panoptes can assist pathologists in determining molecular subtypes and mutations without the necessity of sequencing.
• Further refinement of this AI tool could enhance its utility in personalized treatment strategies for endometrial cancer.