The present study focuses on the use of BioBERT for the prediction of medical diagnoses by processing reports obtained through collaboration with the "Giovanni Paolo II" Cancer Institute in Bari and nine other Italian hospitals. Specifically, a machine learning model is introduced that deals with the extraction and labeling through BIO labeling of relevant information from the reports. So, the BioBERT model is trained to predict patient diagnoses based on this information. The results obtained show that the use of this transformer significantly improves the accuracy of predictions compared with basic models and other natural language processing approaches.
Il presente studio si concentra sull’utilizzo di BioBERT per la predizione di diagnosi mediche attraverso l’elaborazione di report ottenuti attraverso la collaborazione con l’Istituto Tumori “Giovanni Paolo II” di Bari ed altri nove ospedali italiani. In particolare, viene introdotto un modello di machine learning che si occupa dell’estrazione ed etichettatura tramite il labelling BIO delle informazioni rilevanti dei report. Dunque, il modello BioBERT viene addestrato per predire le diagnosi dei pazienti sulla base di tali informazioni. I risultati ottenuti dimostrano che l’utilizzo di questo transformer migliora significativamente l’accuratezza delle predizioni rispetto ai modelli di base e ad altri approcci di elaborazione del linguaggio naturale.
NAMED ENTITY RECOGNITION PER LA DIAGNOSI DI LINFOMA DA REPORTS DI TESTO LIBERO: UN APPROCCIO BASATO SU TRANSFORMER
CALÒ, FEDERICA
2021/2022
Abstract
The present study focuses on the use of BioBERT for the prediction of medical diagnoses by processing reports obtained through collaboration with the "Giovanni Paolo II" Cancer Institute in Bari and nine other Italian hospitals. Specifically, a machine learning model is introduced that deals with the extraction and labeling through BIO labeling of relevant information from the reports. So, the BioBERT model is trained to predict patient diagnoses based on this information. The results obtained show that the use of this transformer significantly improves the accuracy of predictions compared with basic models and other natural language processing approaches.I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14253/81999