Histopathology Images-Based Deep Learning Prediction of Histological Types in Endometrial Cancer.

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence

Background:

• The 2023 FIGO staging system classifies endometrial cancer (EC) into aggressive and non-aggressive types.
• Accurate histological typing of EC is vital for treatment and prognosis.

Purpose of the Study:

• To develop and validate a deep convolutional neural network (CNN) for differentiating aggressive from non-aggressive EC histological types using H&E-stained images.
• To assess the generalizability, clinical utility, and correlation with molecular subtypes of the developed CNN model.

Main Methods:

• A CNN, EC-AI^HIS, was trained on 1187 EC specimens.
• The model's performance was evaluated using cross-validation and external validation cohorts.
• Clinical utility was assessed via simulations, benchmarking against pathologists, and correlation with molecular subtypes.

Main Results:

• EC-AI^HIS achieved an AUC of 0.911, with sensitivity of 82% and specificity of 83%.
• The model demonstrated robustness on varied image quality and scanner types.
• EC-AI^HIS improved junior pathologists' diagnostic accuracy and showed prognostic potential in the p53abn subtype.

Conclusions:

• EC-AI^HIS is an effective AI tool for assisting pathologists in classifying endometrial cancer histological types.
• The model aids in distinguishing aggressive from non-aggressive EC, potentially improving patient management.