Machine Learning for Clinical Literature Review

Mission

Development of an algorithm to extract and select scientific articles for the clinical evaluation of medical devices. Design of an automated pipeline combining web scraping, machine learning and interactive visualization to speed up the preparation of clinical evaluation reports (CE marking). Automated collection of scientific references: extraction, sorting and metadata enrichment from PubMed, Google Scholar, ScienceDirect, Embase. Optimization of the systematic review process. Hybrid (human + ML) classification of relevant studies according to defined inclusion/exclusion criteria (PICO/PICOS, PRISMA…). Integration of automated scoring, a ranking module and optional manual selection to ensure the quality and transparency of the regulatory process. Generation of analytical dashboards and dynamic reporting. Design of a SaaS interface in Python/Dash for the visualization, tracking and auditability of the entire process, compliant with regulatory requirements (MEDDEV 2.7/1, MDCG…). Multi-format export: BibTeX, CSV, XLSX, serving business users (regulatory affairs, consultants, medical device manufacturers). Research & development in artificial intelligence applied to healthcare regulation. Implementation of advanced Natural Language Processing (NLP) methods to extract structured information on device safety and performance. Modular approach: prototyping, iterative testing, benchmarking against the state of the art (Abstrackr, Rayyan, ASReview…).

Contexte & Environnement

Collaboration: Deeptech start-up (WEEKI) + regulatory affairs expert firm (B CONSULTANT), supported by a multidisciplinary university team. "Human-in-the-loop" approach: The user stays in control of the decision; the algorithm assists and explains each step, fostering ownership and trust. Interoperability: Compatibility with healthcare frameworks (PRISMA, PROSPERO), scalable SaaS architecture (microservices).

Objectifs Cles

Technologies & Infrastructure