A carregar...
Publicação
Temporal unsupervised learning models to study ALS progression
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.14 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Amyotrophic Lateral Sclerosis (ALS) is a rapidly progressing chronic disease that affects motor neurons, leading to progressive disability and eventually paralysis. Due to its complexity and
heterogeneity, the search for effective treatments that can slow down the progression of ALS and
improve the quality of life of patients has been a constant challenge in the medical field[54]. Thus,
it is important to automatically identify the groups of patients with similar progressions, to improve the prediction of medical procedures. This work is divided into three parts: stratification,
prognosis of the type of progression, and prediction of the need for medical procedures. In stratification, groups of patients with similar progressions are found to help new patients predict their
needs, resulting in five groups: lower limbs, upper limbs, bulbar, diffuse (with a strong respiratory component), and advanced progressions. Supervised machine-learning models were built to
predict the type of progression, using data collected directly at the first consultation, with some
classifiers showing an accuracy of over 80%. However, it was difficult to predict patients with
diffuse progression, and data balancing techniques were applied which, although they performed
slightly worse overall, showed an improvement in this specific group of patients. The need for
medical procedures varies according to the type of progression. It was calculated which procedures each cluster tends to need and studied whether creating specific classifiers for each of them
would perform better in predicting these procedures when compared to a general classifier. Predictions were made in windows of 90, 180, and 365 days. Comparing the specialized classifiers with
the general ones, inconclusive results were obtained at 90 days, but at 365 days, the specialized
models showed better results in some procedures, such as predicting non-invasive ventilation and
the need for a communication aid device. This project is a step towards a better understanding of
ALS in order to contribute to the development of more personalized therapeutic strategies in the
future.
Descrição
Tese de mestrado, Engenharia Informática, 2023, Universidade de Lisboa, Faculdade de Ciências
Palavras-chave
Esclerose Lateral Amiotrofica Aprendizagem Automática Estratificação Temporal não-supervisionada Procedimentos médicos ELA Teses de mestrado - 2024