Related Experiment Video
Updated: Sep 5, 2025
07:35
A Knowledge Graph Approach to Elucidate the Role of Organellar Pathways in Disease via Biomedical Reports
Published on: October 13, 2023
1.7K
Exploring deep learning methods for recognizing rare diseases and their clinical manifestations from texts.
Isabel Segura-Bedmar1, David Camino-Perdones2, Sara Guerrero-Aspizua3,4,5,6
1Human Language and Accesibility Technologies, Computer Science Department, Universidad Carlos III de Madrid, Avenidad de la Universidad, 30, Leganés, 28911, Madrid, Spain. isegura@inf.uc3m.es.
BMC Bioinformatics
|July 6, 2022
Summary
Deep learning models, including BioBERT, show promise in identifying rare diseases and their symptoms. While BioBERT achieved an 85.2% F1 score for rare disease recognition, symptom identification requires further improvement.
Area of Science:
- Medical Informatics
- Computational Linguistics
- Rare Disease Research
Background:
- Rare diseases affect 400 million people globally, posing diagnostic challenges for general practitioners due to varied manifestations and limited knowledge.
- Delayed diagnosis of rare diseases significantly impacts patient outcomes, highlighting the need for enhanced medical understanding and diagnostic tools.
Purpose of the Study:
- To explore the application of Natural Language Processing (NLP) and Deep Learning techniques for the early and accurate diagnosis of rare diseases.
- To assess the effectiveness of advanced deep learning models in recognizing rare diseases and their associated clinical manifestations (signs and symptoms).
Main Methods:
- Investigated deep learning models, including Bidirectional Long Short Term Memory (BiLSTM) networks and Bidirectional Encoder Representations from Transformers (BERT).
- Utilized BioBERT, a BERT-based model trained on biomedical corpora, for recognizing rare diseases and their clinical features.
Main Results:
- BioBERT achieved a high F1 score of 85.2% in recognizing rare diseases.
- The model's performance in identifying clinical manifestations (signs and symptoms) was lower, with an F1 score of 57.2%, due to the complexity of medical descriptions.
Conclusions:
- The study demonstrates the potential of deep learning, particularly BioBERT, in aiding rare disease diagnosis.
- Further research and model refinement are necessary to improve the accurate identification of complex clinical manifestations for rare diseases.
