Automated Digitization of Paper-based Medical Records

Introducing elderly into specific continence products is today done by analog tracking workflows. In current workflows, caregivers within the elderly health care manually record events such as bathroom visits and leakages on paper and then manually transfer it to digital data, a process that consumes time and is prone to errors due to the human factor. This thesis addresses these challenges by investigating the use of Optical Character Recognition (OCR) and computer vision techniques to automate the extraction and structuring of data from images of these records. This project is a design science based study and was conducted in collaboration with Framna and supported by Linnaeus University. Building on existing research in OCR, image processing and machine learning-based document analysis, the system is designed to process semi-structured medical records, often handwritten and inconsistently formatted, captured via smartphone cameras. A prototype was developed using pre-trained tools and evaluated on a dataset of digital and photographed real-world records. Results were measured in terms of accuracy, efficiency, and robustness across varying conditions. The findings demonstrate the feasibility of such a system in reducing manual workload and increasing documentation accuracy, while also highlighting integration challenges in healthcare environments. The project contributes a domain-specific application of OCR in medical documentation, emphasizing efficiency, and workflow compatibility.