HistoNav: AI-Based H&E Histopathology Predicts Disease-Specific Survival and Guides Adjuvant Chemotherapy Decisions
Xianhong Xu1, Susan Fotheringham1, Surya Rajan1
1Oxford Cancer Biomarkers, The Magdalen Centre, Oxford Science Park, Robert Robinson Avenue, Oxford OX4 4GA, UK.
Abstract:
Background: Stage II/IIIA colorectal cancer (CRC) patients have a relatively high 5-year overall survival rate after surgical resection alone, but more than 50% of patients receive adjuvant chemotherapy. This study investigates a technology that applies artificial intelligence (AI) to conventional histopathology images to identify patients at risk of recurrence. Methods: HistoNav, a novel AI deep-learning algorithm based on Vision Transformer, Graph Neural Networks and Convolutional Neural Networks was designed to analyse H&E-stained formalin-fixed paraffin-embedded (FFPE) samples (n = 2095) to stratify Stage II/IIIA CRC patients into low-, intermediate-, and high-risk groups. Results: Data analysis revealed a 5-year disease-specific survival (DSS) of 93.2%, 84.3%, and 69.3% for low-, intermediate-, and high-risk groups, respectively. The hazard ratio for the high versus low-risk group (HR = 4.611, 95% CI: 2.776-7.662; p < 0.000001) was statistically significant, demonstrating HistoNav's potential to stratify patients based on recurrence risk. Conclusions: HistoNav can effectively identify CRC patients with good prognosis using the digital images of H&E-stained resection samples and will support clinical decisions around the use of adjuvant chemotherapy.
