Development of a Symptom-Based GI Cancer Diagnostic Bot Using BioBERT-NLI, FLAN-T5 and RAG Model

Gastrointestinal(GI) cancers, which include cancers of the stomach, pancreas, liver and intestines, are among the major contributors to cancer-related deaths worldwide. Given the non-specific nature of some earlier symptoms, the early diagnosis has remained a key yet difficult goal. This paper is aimed at creating a symptom-based diagnostic bot to support clinicians in timely detection and planning for treatment of GI cancer in line with the critical need for efficient diagnostic tools. The method integrates three advanced natural language processing (NLP) components: BioBERT-NLI encode biomedical text into contextual embeddings, retrieval-augmented generation leverage FAISS to retrieve related medical literature and FLAN-T5 to generate coherent evidence-based outputs. The bot was trained and evaluated on a dataset of 13,000 GI cancer-related question-answer pairs from PubMedQA. Performance was evaluated on 70 clinically staged query-response pairs. The results showed solid semantic fidelity as reflected by BERTScore F1 85.21, though constraints were colorful concerning phrasing (BLEU: 2.62) and extraction of entities (F1: 21.05). The bot becomes a clinical decision-support tool latently, enabling doctors to process vast sheets of biomedical literature to boost diagnostic accuracy. Enhancements in the future will focus on widening the training dataset while further refining the responses and entity recognition of the improvements.