Predicting Intratumoral Heterogeneity and Stratifying Prognostic Risk in Gastric Cancer Using a Histology-Based Pathomics-Deep Learning Fusion Model.

Abstract

Intratumoral heterogeneity (ITH) is a fundamental driver of clonal evolution and therapeutic resistance in gastric cancer (GC). However, the clinical assessment of ITH remains limited by the high cost and technical complexity of multi-region and single-cell sequencing. This study aimed to develop a pathomics-deep learning (DL) fusion model for estimating ITH directly from routine H&E-stained whole-slide images (WSIs) in GC. We retrospectively collected 893 WSIs from 773 patients in three independent cohorts. ITH was quantified using eight algorithms, and the optimal prognostic indicator was identified by Cox regression analysis. Pathomics features were extracted using CellProfiler, whereas DL features were generated using ResNet50 combined with two multiple-instance learning pipelines. A pathology-driven stacking ensemble (PASTE) model integrating pathomics and DL features was then constructed. A total of 239, 103, 135, and 30 patients were included in the training, internal validation, and two external test cohorts, respectively. The mutant-allele tumor heterogeneity (MATH) score was identified as an independent prognostic factor for overall survival (OS) (HR, 1.840; 95% CI, 1.144-2.957; P = 0.012). Biological relevance analysis showed that high-MATH tumors were characterized by increased chromosomal instability and an immunosuppressive microenvironment, whereas low-MATH tumors were associated with immune-active phenotypes. The PASTE model showed favorable performance in predicting MATH-defined ITH status, with AUCs of 0.852-0.956 across the training, validation, and test cohorts. Moreover, the model-derived ITH risk score was an independent prognostic factor and effectively stratified patients into high- and low-risk groups with significantly distinct OS outcomes (all P < 0.05). Further interpretability analysis of the DL component using Grad-CAM showed that the model primarily attended to regions characterized by nuclear atypia and immune-cell infiltration. In conclusion, we developed a robust and interpretable H&E-based model for predicting MATH-defined ITH and supporting prognostic stratification in patients with GC. This AI-driven approach may provide a cost-effective and scalable tool to support precision oncology in clinical practice.