Herramienta digital basada en inteligencia artificial para la extracción y verificación automatizada de información contenida en documentos clínicos escaneados

Abstract

Medical disability management in Colombia faces serious structural difficulties that impact the traceability and reliability of information. Today, manual transcription and verification processes are slow, costly, and also prone to human error or even document fraud. All of this creates operational overloads that affect both healthcare institutions and companies. Faced with this situation, the need to create a system that automates the workflow and offers security, transparency, and trust in the validation of these documents is evident. Therefore, the purpose of this project is to implement a digital tool, supported by artificial intelligence, that allows for the automatic extraction and verification of information contained in scanned clinical documents. To achieve this, the development of a web application is proposed, using advanced AI models capable of interpreting images without relying on traditional OCR. In this way, the data obtained is structured and automatically cross-referenced with official sources from the health system. The solution combines a React frontend, a FastAPI backend, data management in Supabase, and validation based on clinical and regulatory criteria. With the development of this AI-based digital tool, the process of transcribing and verifying scanned medical disability certificates was automated, allowing the extraction of key fields (patient data, ICD-10 diagnosis, disability dates, and issuing physician) directly from images, structuring the information, and validating its authenticity through queries to official sources of the Colombian health system. This reduces processing times compared to manual methods, decreases transcription errors, and strengthens control against document fraud, benefiting companies, EPS (Health Promoting Entities), ARL (Occupational Risk Administrators), and employers by optimizing operational resources and reducing costs associated with unjustified absenteeism. The validation of the prototype was carried out through tests with 39 simulated and anonymized scanned medical disability certificates, measuring the average processing time per document compared to the manual method, the accuracy of information extraction, and the effectiveness of the verification process, including the success of queries to official portals and the detection of inconsistencies. The results obtained confirm the reduction in times and errors in the evaluated set. In this way, the community benefits by achieving greater speed in the recognition of labor rights and economic benefits for disabilities, by reducing delays and unjustified rejections that affect workers and patients. Furthermore, transparency and sustainability in the Colombian health system are promoted by mitigating fraud that impacts public and private resources, and more efficient administrative processes in institutions are facilitated, contributing to a more reliable management of sensitive clinical information.