Chronic pain classificationa machine learning perspective

Laburpena

Chronic Pain is a complex multifactorial problem that affects 1 in 5 Europeans with the great majority suffering from pain for 7 years or more and 34% of them describing the pain experience as severe. As a consequence, chronic pain patients have their quality of life deteriorated and their chances of unemployment increased. An early diagnosis and treatment of chronic pain can improve the quality of life and reduce the recovery time. Thus, primary care has a central role as 70% of patients with chronic pain are managed in primary care. On the other side, primary care professionals have demonstrated inappropriate attitudes and beliefs about pain and its treatment even after participating in continuous educational programs. As a consequence, primary care physicians do not feel confident to provide a chronic pain diagnosis. This may affect the time a patient has to wait for a suitable diagnosis with 20.9% of patients waiting for more than 10 years. Thus, primary care professionals and patients could benefit from objective methods to support the diagnosis decision. In the last decade, machine learning has been widely used in different fields, especially because of its capacity to work with complex data. With the support of machine learning techniques, different studies have been using data-driven approaches to better understand some syndromes like mild cognitive impairment, Alzheimer's disease, and schizophrenia. In this thesis we assessed the sensitivity of different algorithms and datasets diagnosing chronic pain syndromes. Together with this assessment, we highlighted important methodological steps that should be taken into account when an experiment using machine learning is conducted. As a result, we are able to diagnose chronic pain with a fair to good accuracy using both self-reported and subjective data (questionnaire and QST) as well as more objective data from resting-state functional magnetic resonance images. Also, analyzing brain images, we observed that a convolutional neural network outperformed less costly and more commonly adopted models such as support vector machines (SVM) and RFC. On the other hand, using questionnaires and QST, we observed that ensemble-based methods show the best performance. Additionally, the performance of an algorithm is strongly related to the hyper-parameters and a good strategy for hyper-parameter optimization should be used to extract the most from the algorithms. Finally, the most important contribution of this thesis is an explanation for machine learning predictions diagnosing chronic pain. That explanation highlights that algorithms make their decision based on scientific evidence existing in the literature. These explanations are important to increase the trust of patients and professionals about machine predictions. This thesis supports the notion that machine learning can be a powerful tool to better diagnose chronic pain conditions and to augment the abilities of health care professionals to manage chronic pain.