Medical document digitization and intelligent information extraction represent foundational challenges in automating health insurance workflows. Existing automated claim processing approaches predominantly treat document reading as a preprocessing step, devoting limited attention to the accuracy, robustness, and structured information extraction capabilities of the underlying document intelligence pipeline. This paper addresses this gap by presenting a medical document intelligence system that places OCR-driven bill analysis at the center of an automated insurance claim processing architecture. The proposed system employs EasyOCR, a deep learning text recognition framework combining CRAFT region detection with ResNet-LSTM sequence decoding, to digitize uploaded medical bills in PDF and image formats. A structured post-processing pipeline extracts seven categories of clinical and financial information from raw OCR output including hospital identity, patient demographics, diagnostic codes, itemized charges, subtotal, discount, and net payable amount. Extracted information feeds a semantic policy matching engine that combines FAISS-indexed sentence embeddings with the Groq LLaMA 3.3 70B large language model to evaluate claims against a vector-indexed insurance policy handbook. A bill amount cross-verification mechanism detects inflated claim submissions by comparing the declared claim amount against the OCR-extracted net payable figure. The complete system is implemented in Python using the Flask web framework and deployed on Hugging Face Spaces. Evaluation on a test corpus of eight medically diverse claim scenarios achieves 100% correct verdict assignment. OCR pipeline evaluation on a printed medical bill photograph demonstrates successful extraction of all seven target information categories including correct identification of the net payable amount for fraud cross-verification.
Paper
Full text
Automated Medical Document Intelligence for Health Insurance Processing
Semantic Scholar · 2026
Abstract
Medical document digitization and intelligent information extraction represent foundational challenges in automating health insurance workflows. Existing automated claim processing approaches predominantly treat document reading as a preprocessing step, devoting limited attention to the accuracy, robustness, and structured information extraction capabilities of the underlying document intelligence pipeline. This paper addresses this gap by presenting a medical document intelligence system that places OCR-driven bill analysis at the center of an automated insurance claim processing architecture. The proposed system employs EasyOCR, a deep learning text recognition framework combining CRAFT region detection with ResNet-LSTM sequence decoding, to digitize uploaded medical bills in PDF and image formats. A structured post-processing pipeline extracts seven categories of clinical and financial information from raw OCR output including hospital identity, patient demographics, diagnostic codes, itemized charges, subtotal, discount, and net payable amount. Extracted information feeds a semantic policy matching engine that combines FAISS-indexed sentence embeddings with the Groq LLaMA 3.3 70B large language model to evaluate claims against a vector-indexed insurance policy handbook. A bill amount cross-verification mechanism detects inflated claim submissions by comparing the declared claim amount against the OCR-extracted net payable figure. The complete system is implemented in Python using the Flask web framework and deployed on Hugging Face Spaces. Evaluation on a test corpus of eight medically diverse claim scenarios achieves 100% correct verdict assignment. OCR pipeline evaluation on a printed medical bill photograph demonstrates successful extraction of all seven target information categories including correct identification of the net payable amount for fraud cross-verification.