A Novel Model for Predicting Microsatellite Instability in Endometrial Cancer: Integrating Deep Learning-Pathomics...

Area of Science:

• Oncology
• Radiology
• Pathology

Background:

• Microsatellite instability (MSI) is a crucial biomarker in endometrial cancer (EC).
• Accurate prediction of MSI status is essential for guiding treatment decisions in EC.
• Current methods for MSI detection can be invasive or time-consuming.

Purpose of the Study:

• To develop and validate a novel deep learning pathoradiomics model (DLPRM).
• To predict microsatellite instability (MSI) status in endometrial cancer (EC) patients.
• To utilize multiparametric MRI (mpMRI) and whole slide images (WSIs) for non-invasive MSI prediction.

Main Methods:

• A retrospective study included 136 EC patients, divided into training (96) and validation (40) sets.
• Deep learning (ResNet50) extracted pathomics features; Pyradiomics extracted radiomics from T2WI, DWI, and AP sequences.
• A multilayer perceptron (MLP) integrated radiomics and pathomics features into the DLPRM, validated using ROC, accuracy, sensitivity, specificity, PPV, NPV, and F1-score.

Main Results:

• A DLPRM combining radiomics and pathomics signatures demonstrated high performance in predicting MSI status.
• The DLPRM achieved an AUC of 0.960 in the training set and 0.917 in the validation set.
• Decision curve analysis indicated significant clinical net benefits of the DLPRM.

Conclusions:

• The DLPRM effectively predicts MSI status in EC patients using pretreatment pathoradiomics images.
• The model exhibits high accuracy and robustness, offering a novel tool for individualized EC management.
• This approach could assist clinicians in optimizing treatment strategies for EC patients.