Development of machine learning and artificial intelligence techniques in the study of cancer-related cachexia

Author(s)

Natasha Fioretto Aguero

• Document type: Doctoral Thesis
• Press: São Paulo.
• Institution: Universidade de São Paulo (USP). Instituto de Física (IF/SBI)
• Defense date: 2023-05-15

Examining board members

Alexandre Alarcon do Passo Suaide; Gustavo Martini Dalpian; Amauri Jardim de Paula; Marilia Cerqueira Leite Seelaender; Elisabeth Mateus Yoshimura

• Advisor: Alexandre Alarcon do Passo Suaide

Abstract

With the aging trend of the world population, public health issues related to older adults, such as cancer incidence, become increasingly important. Cancer patients may develop a syndrome called cachexia, in which there is a tremendous loss of skeletal muscle mass that, when detected, has little chance of cure with current diagnoses. Although artificial intelligence techniques have been successfully applied to study different cancer types, cachexia remains ignored in this field. In this work, we used biochemical data from cachectic and non-cachectic patients in an unsupervised learning algorithm (\\textit{k-means clustering}) in order to test how such a method would group these individuals, which parameters would be of greater importance and the possibility of diagnosing the pre-cachexia phase. We observed that specific clinical parameters, such as high CRP and low cholesterol, were common among patients classified as cachectic by the algorithm, even when there was no significant weight loss. Still, on human aging, we also studied the feasibility of these algorithms in the study of biomechanical predictors of falls, one of the most common external causes of death among the elderly. For this purpose, we analyzed a public database with questionnaire responses and dynamic force balance tests. Our goal was to observe clusters formed with unsupervised algorithms and the applicability of supervised algorithms. Obtaining a behavior that could define individuals with a tendency to fall was found to be challenging. There was little differentiation of groups in questionnaire responses, but we could distinguish between age groups in tests on strength balances. (AU)

• FAPESP's process: 17/17096-1 - Development of machine learning techniques and artificial intelligence for the study of cancer related cachexia
• Grantee: Natasha Fioretto Aguero
• Support Opportunities: Scholarships in Brazil - Doctorate