Interpretable multi-modal artificial intelligence model for predicting gastric cancer response to neoadjuvant...

Area of Science:

• Oncology
• Radiology
• Pathology
• Artificial Intelligence

Background:

• Accurate assessment of neoadjuvant chemotherapy response is crucial for locally advanced gastric cancer (LAGC) prognostication and treatment planning.
• Current methods for predicting treatment efficacy have limitations, necessitating advanced predictive tools.

Purpose of the Study:

• To develop and validate an interpretable artificial intelligence (AI) framework, the incremental supervised contrastive learning model (iSCLM), for predicting neoadjuvant chemotherapy response in LAGC.
• To integrate pretreatment computed tomography (CT) scans and hematoxylin and eosin (H&E)-stained biopsy images for enhanced predictive accuracy.

Main Methods:

• The iSCLM framework was developed using retrospective data from 2,387 LAGC patients across 10 medical centers.
• Model performance was evaluated on a prospective cohort of 132 patients (ChiCTR2300068917).
• Interpretable AI techniques, including Shapley additive explanations and global sort pooling, were used to generate attention heatmaps for CT and pathology images.

Main Results:

• The iSCLM achieved high discriminative ability with areas under the receiver operating characteristic curves ranging from 0.846 to 0.876 across test cohorts.
• Attention heatmaps highlighted the model's ability to capture relevant morphological features from both CT and pathology data.
• Pathological analysis revealed that top-ranked tiles in responders showed decreased distance to tumor-invasive borders and increased inflammatory cell infiltration compared to non-responders.
• Elevated CD11c expression was observed in responders, indicating a potential molecular biomarker.

Conclusions:

• The developed interpretable AI model (iSCLM) accurately predicts chemotherapy efficacy in LAGC patients at the molecular pathology level.
• iSCLM offers a promising tool for improving clinical decision-making and patient management regarding neoadjuvant chemotherapy.
• The integration of imaging and histopathology data through AI enhances the understanding of treatment response mechanisms.