PROJECTS
PROJECTS
For ASA Laser , on behalf of a client, I developed an automatic ICD-9 code classification pipeline. Since the process is usually complex and resource-intensive, I decided to use GPT-4o to evaluate whether it could represent the fastest and most efficient solution to automate this task.
The ICD-9 (International Classification of Diseases, 9th Revision) is a standardized coding system widely used for clinical documentation, hospital billing, epidemiological studies, and medical research.
The project followed three main steps:
Model validation – GPT-4o was benchmarked on the open MIMIC-III-50 dataset, ensuring reliability through precision and F1 metrics.
Proof of concept – The script was tested on a public database to refine the methodology and assess scalability.
Deployment on real data – Finally, the pipeline was applied to a client dataset of over 200,000 patient records, automatically classifying diagnoses and extracting structured information such as anatomical region and laterality.
This project showed that large language models can be an effective tool for accelerating medical data processing, providing clinicians and researchers with faster, structured insights. In particular, the model did not only assign the ICD-9 code but also extracted additional fields such as Anatomical Region and Laterality, making the dataset richer and more useful for clinical and research applications.
For further information about the repository, please contact me.
For my Master’s thesis at EPFL (Lausanne, 2023), I worked on the development of a prototype device designed to monitor the human body’s hydration status. The project focused on the characterization of flexible, non-invasive sensors, including:
a temperature sensor (RTD) to detect body temperature changes,
an interdigitated sensor to measure local skin hydration,
tetrapolar electrodes for Bio-Impedance Analysis (BIA) to assess total body water content.
The sensors were validated both in vitro and in vivo, confirming their ability to detect variations in hydration. These results represent a solid proof of concept for future wearable, low-cost, and stand-alone medical devices that could help prevent dehydration, particularly in elderly patients.
Link to the full thesis: Master Thesis