“Performance Comparison of Fine-tuned BERT-Based Genomic Language Models” è il quarto webinar scientifico che si svolgerà il 14 marzo dalle ore 12 alle ore 13 a cura Payel Patra dell’Università de L’Aquila. Patra è dottoranda al terzo anno del dottorato di ricerca in ICT dell’Università dell’Aquila con borsa finanziata dal progetto SoBigData.it e lavora su tematiche di analisi di dati medici e clinici attraverso approcci di AI.
Abstract
The evolution of articial intelligence can empower accurate medical diagnostics (e.g., disease, Symptoms, and treatment), personalized treatment plans, and medical data analysis, speeding up the diagnosis or improving treatments. As a primary study, we conducted a comparison among several BERT-based models ne-tuned on unstructured Clinical Summary Mammary Malignancy (CSMM) data to annotate clinical notes with breast cancer-specic concepts. Our research transforms unstructured clinical text data into a structured format (CSV), simplifying the identication and recognition of specic entities (namely, Clinical Named Entity Recognition -NER), such as age, gender, disease, symptoms, medication, doses, medical history, and cancer stage. The considered BERT-based models are BioBERT, BioClinicalBERT, RoBERTa, PubMedBERT and BlueBERT. The comparative analysis reveals that our ne-tuned BioBERT transformer model demonstrates more robust performance, achieving a high F-score of 96% in extracting medical entities, which conrms their e‑ectiveness in handling domain-specic clinical texts.
Per seguire il seminario, ci si può collegare al link: https://meet.google.com/vaa-oczm-yjy
