Integrating TCM Theory with AI: A BP-GA Hybrid Framework for Precision Optimization of Multi-Herb Formulations in Gastric Cancer Treatment

Abstract

Gastric cancer remains a leading cause of global cancer mortality, necessitating innovative therapeutic strategies that balance efficacy with safety. This study presents a novel hybrid artificial intelligence framework integrating backpropagation neural networks (BPNN) and genetic algorithms (GA) to optimize dosing strategies for a multi-herb Traditional Chinese Medicine (TCM) prescription in gastric cancer treatment. Our approach uniquely addresses three critical challenges: (1) the nonlinear pharmacodynamics of herbal combinations, (2) the quantitative operationalization of classical TCM theory (Jun-Chen-Zuo-Shi principles), and (3) the reconciliation of empirical knowledge with precision medicine requirements. Through Plackett-Burman screening and orthogonal experimental designs, we identified Actinidia arguta (28.7% contribution, p=0.003), Scutellaria barbata (22.4%, p=0.007), and Strychnos nux-vomica (18.9%, p=0.012) as key bioactive components, demonstrating significant synergistic interactions (?-inhibition >15%, p<0.001). The BP-GA model achieved superior predictive accuracy (R²=0.941) compared to conventional methods, enabling precise dosage optimization via Pareto front analysis (63g Actinidia arguta, 68g Scutellaria barbata, 4.8g Strychnos nux-vomica). In a phase III randomized controlled trial (n=216), the optimized formulation significantly extended median survival by 4.3 months (HR=0.62, p<0.001), improved response rates (82.4% vs 68.7%, p<0.05), and reduced gastrointestinal toxicity by 35.7% (p<0.05). These results validate our computational approach in bridging TCM theory with clinical outcomes while establishing a new paradigm for evidence-based herbal medicine development. The framework's adaptability suggests broad applicability across integrative oncology and complex polypharmacy optimization.