Publicação
Using Machine Learning to Predict Mobility Improvement of Patients after Therapy: A Case Study on Rare Diseases
| Resumo: | A disease is considered rare when it affects less than 5 people in 10 000 in the European Union. Although each disease individually affects a low number of patients, it is estimated to exist approximately 350 million people worldwide struggling with rare diseases. Despite these conditions threaten a significant part of the population, the cause, treatment and possible cure for many of them is usually unknown. Given the lack of information, investigation, and support around this topic, the number of studies regarding rare diseases is still scarce, and even less so the ones that use Machine Learning to support them. With this in mind, the present research aims to develop a Machine Learning model capable of assisting medical decisions in this field. There are many manifestations and symptoms associated with rare diseases, but a common consequence is brain cell degeneration, which in turn causes the deterioration of patients' motor capabilities. The treatment process for these patients can involve several physiotherapy sessions, which imply high costs and drawbacks. For these reasons, it is becoming more and more important to distinguish a respondent patient from a non-respondent. In this way, the principal goal of this research was to develop a Machine Learning model capable of predicting the improvement of the motor functions of patients after attending physiotherapy, determining whether they will respond or not to the treatment. These predictions provide a base for planning the clinical treatment and assisting in medical decisions. Artificial Intelligence and Machine Learning can be particularly beneficial in rare disease scenarios, given their ability to uncover complex data patterns and memorize data as no human being can. However, the lack of data is the main drawback for both rare disease case studies and Machine Learning models. In this way, Data Augmentation techniques were further explored to generate more data, enhancing the conditions to make use of Machine Learning algorithms. Besides that, this research resorted to several types of predictive models, such as the Regularization models, which are well known for their capability of generalizing new data. Generalization is an essential characteristic when dealing with such unique and different diseases. Besides that, ensembles of models were also considered since Ensembles boost the learning of individual algorithms and avoid overfitting. Ensembles turned out to be the best model to predict the overall patients' motor functions after the therapy, with an error rate of approximately 4%. |
|---|---|
| Autores principais: | Oliveira, Lara Barradas Teixeira Garrucho de |
| Assunto: | Machine learning Rare Diseases Data Augmentation Data Mining Small Data Imbalanced Regression |
| Ano: | 2023 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | A disease is considered rare when it affects less than 5 people in 10 000 in the European Union. Although each disease individually affects a low number of patients, it is estimated to exist approximately 350 million people worldwide struggling with rare diseases. Despite these conditions threaten a significant part of the population, the cause, treatment and possible cure for many of them is usually unknown. Given the lack of information, investigation, and support around this topic, the number of studies regarding rare diseases is still scarce, and even less so the ones that use Machine Learning to support them. With this in mind, the present research aims to develop a Machine Learning model capable of assisting medical decisions in this field. There are many manifestations and symptoms associated with rare diseases, but a common consequence is brain cell degeneration, which in turn causes the deterioration of patients' motor capabilities. The treatment process for these patients can involve several physiotherapy sessions, which imply high costs and drawbacks. For these reasons, it is becoming more and more important to distinguish a respondent patient from a non-respondent. In this way, the principal goal of this research was to develop a Machine Learning model capable of predicting the improvement of the motor functions of patients after attending physiotherapy, determining whether they will respond or not to the treatment. These predictions provide a base for planning the clinical treatment and assisting in medical decisions. Artificial Intelligence and Machine Learning can be particularly beneficial in rare disease scenarios, given their ability to uncover complex data patterns and memorize data as no human being can. However, the lack of data is the main drawback for both rare disease case studies and Machine Learning models. In this way, Data Augmentation techniques were further explored to generate more data, enhancing the conditions to make use of Machine Learning algorithms. Besides that, this research resorted to several types of predictive models, such as the Regularization models, which are well known for their capability of generalizing new data. Generalization is an essential characteristic when dealing with such unique and different diseases. Besides that, ensembles of models were also considered since Ensembles boost the learning of individual algorithms and avoid overfitting. Ensembles turned out to be the best model to predict the overall patients' motor functions after the therapy, with an error rate of approximately 4%. |
|---|