2024: Testing AI and ML tools for structuring unstructured medical texts

The project tackled the challenge of processing unstructured Estonian clinical texts, which are resource-intensive and lack machine-readability, hindering healthcare efficiency. The aim was to test an AI tool that automatically structures medical free text and maps it to SNOMED CT codes to reduce administrative burden. Natural Language Processing (NLP) technologies, primarily Microsoft Azure Text Analytics for Health, were validated. A web-based prototype was tested to perform named entity recognition, relationship extraction, and generate FHIR-compliant outputs. The prototype was tested on general practitioner notes and radiology reports, with three doctors annotating the texts. It successfully identified clinical terms and mapped them to SNOMED CT, though challenges persisted with Estonian morphological variations. The solution reduced manual processing and improved data quality. Future plans include model fine-tuning and integration with Estonia’s health information system. The project supports time savings and enhanced diagnostic accuracy in healthcare.