Related Experiment Videos
Predicting Intratumoral Heterogeneity and Stratifying Prognostic Risk in Gastric Cancer Using a Histology-Based
Shulun Nie1, Duanbo Shi2, Shuyi Song1
1Department of Medical Oncology, Qilu Hospital, Cheeloo College of Medicine, Shandong University, Jinan, Shandong, China.
Summary
This study developed an AI model using routine H&E slides to predict intratumoral heterogeneity (ITH) in gastric cancer. The model accurately estimates ITH, aiding in patient prognosis and potentially improving precision oncology.
Area of Science:
- Oncology
- Computational Pathology
- Artificial Intelligence
Background:
- Intratumoral heterogeneity (ITH) is a key factor in gastric cancer (GC) progression and treatment resistance.
- Current methods for assessing ITH, like multi-region and single-cell sequencing, are costly and complex.
- There is a need for accessible methods to evaluate ITH in clinical settings.
Purpose of the Study:
- To develop and validate a pathomics-deep learning (DL) fusion model for estimating ITH directly from H&E-stained whole-slide images (WSIs) in GC.
- To assess the prognostic value of the developed model in stratifying GC patients.
- To investigate the biological underpinnings of ITH in GC using AI-driven insights.
Main Methods:
- Retrospective analysis of 893 WSIs from 773 GC patients across three independent cohorts.
- Quantification of ITH using eight algorithms, with the mutant-allele tumor heterogeneity (MATH) score identified as a prognostic indicator.
- Extraction of pathomics features (CellProfiler) and DL features (ResNet50 with multiple-instance learning).
- Construction of a pathology-driven stacking ensemble (PASTE) model integrating pathomics and DL features.
Main Results:
- The MATH score was confirmed as an independent prognostic factor for overall survival (OS) in GC patients.
- High-MATH tumors exhibited increased chromosomal instability and an immunosuppressive microenvironment.
- The PASTE model demonstrated high performance (AUCs 0.852-0.956) in predicting MATH-defined ITH status across all cohorts.
- The model-derived ITH risk score independently predicted OS and stratified patients into distinct risk groups.
Conclusions:
- A robust, interpretable, H&E-based AI model (PASTE) was developed for predicting MATH-defined ITH in GC.
- This AI-driven approach offers a cost-effective and scalable tool for prognostic stratification.
- The model has the potential to enhance precision oncology by providing accessible ITH assessment in clinical practice.