Tools
Tools
Uppalapati, Karthik (2025) MedPredict-DeBERTa: Enhancing Symptom-Based Disease Prediction through UMLS-Integrated Disentangled Attention. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (1MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (1MB) | Preview |
Abstract
The issue of accurate prediction of diseases through symptoms continues to be a major challenge in healthcare. This paper explores how Unified Medical Language System (UMLS) can be integrated with DeBERTa transformer architecture to enhance prediction of diseases according to symptoms of patients. An inclusive pipeline was designed based on the DDXPlus data comprising of 1,025,602 patient records of 49 diseases. This pipeline incorporates UMLS attributes into symptom text including 140 identifiers of concepts, 38 semantic types and 27 critical medical qualifiers. Five models at the baseline and three variants of UMLS integration based on concatenation, gated fusion and cross-attention mechanisms were tested. The findings indicate that the baseline has over 97 percent accuracy. The distilbert variant incorporated into the UMLS has 97.60, which is higher than the DeBertAt baseline of 97.47. The research found the meaning of shallow integration problem, where large performing baselines fail to utilize additional knowledge characteristics completely. Other notable works include a UMLS extraction pipeline, which maintains qualifiers, a comparative study of integration architectures in medical knowledge, and defining under which circumstances medical knowledge integration is valuable. The results indicate that traditional architectural constraints make the use of the knowledge that exists outside to be more successful on small models, and that architectural restrictions are required to make the domain knowledge useful to the systems with the best performance.
| Item Type: | Thesis (Masters) |
|---|---|
| Supervisors: | Name Email Sahni, Anu UNSPECIFIED |
| Subjects: | Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence P Language and Literature > P Philology. Linguistics > Computational linguistics. Natural language processing R Medicine > Diseases R Medicine > Healthcare Industry Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning |
| Divisions: | School of Computing > Master of Science in Artificial Intelligence |
| Depositing User: | Ciara O'Brien |
| Date Deposited: | 02 Sep 2026 11:03 |
| Last Modified: | 02 Sep 2026 11:03 |
| URI: | https://norma.ncirl.ie/id/eprint/9771 |
Actions (login required)
 |
View Item |