Current students - PhD DADS

GIBELLO RICCARDO

Cycle: XXXIX

Advisor: CAIANI ENRICO GIANLUCA
Tutor: MATTEUCCI MATTEO

Major Research topic:
Exploiting Semantic Embeddings and Classification Algorithms to Enhance Medical Device Nomenclature Harmonization and Streamline Regulatory Processes

Abstract:

This project aims to create tools that could streamline and automate vigilance processes for medical devices, following the guidelines outlined by the "CUN" in "Area 09 - Industrial and Information Engineering". It addresses critical needs within regulatory procedures for medical devices, particularly concerning their ongoing monitoring once they are on the market. One significant challenge is the lack of a standardized nomenclature for identifying medical devices, which is particularly evident in Post-Market Surveillance tasks.
Thus, our research endeavours to explore, define, and validate text analytics methodologies for implementing an automated mapping system between the Global Medical Device Nomenclature (GMDN), used within the FDA jurisdiction, and the European Medical Device Nomenclature (EMDN), with potential applicability to other contexts and standards.
This initiative aligns seamlessly with WHO directives urging the adoption of EMDN as a global standard, given its structurally advantageous characteristics. Another important task to be addressed in this research is the maintenance of these nomenclatures. This requires that, for each nomenclature code, its description is aligned and updated with reference to the detailed explanations of how each related device is utilized.
The project unfolds across three key phases, with invaluable input from industry professionals for validation:
;

In the initial phase, we undertake the task of cross-referencing medical device data from both European and American markets, capturing a set of correspondences between GMDN and EMDN codes. These annotated mappings serve as the gold standard for establishing correlations between the two nomenclatures and as a benchmark for subsequent algorithmic training.
This second phase focuses on exploring Natural Language Processing (NLP) methodologies to address two key objectives: mapping GMDN to EMDN nomenclatures and aligning nomenclature code descriptions with related device usages. For the latter, we will delve into document embedding techniques using language models tailored for medical device descriptions. By clustering these documents, we could spot outliers, possibly suggesting device reclassification or the introduction of new nomenclature codes to accommodate emerging technologies. As for the GMDN-EMDN mapping task, we will develop a multi-label hierarchical classification algorithm. This involves partitioning the GMDN-EMDN correspondence dataset into training, validation, and test sets. Leveraging the hierarchical structure of EMDN and exploiting pre-trained biomedical Language Models (e.g., LLMs) to embed GMDN Term Names, we will fine-tune the classification algorithm for accurate mapping.
The validation stage will involve the adoption by professionals of the tools of point (2) to assist their everyday duties, related to the manual mapping between nomenclatures and their maintenance. Through real-world testing scenarios, we would assess the algorithms' performances, thereby applying potential refinements or alternative methodologies to optimize their efficacy.

; Overall, these solutions could be used by regulatory authorities for monitoring purposes. Indeed, the real-time mapping algorithm could be integrated into a platform for Post-Market Surveillance across Europe and beyond. Thanks to this algorithm, the platform would harmonize information by reclassifying devices encoded in different standards into the EMDN framework, facilitating data aggregation and in-depth analysis. Moreover, the implemented solutions could provide services to businesses, streamlining and systematizing the preparation of the documentation required for certification under the European Medical Device Regulation.