ARTIFICIAL INTELLIGENCE FOR EVALUATING GAIT IN CANCER PATIENTS UNDERGOING CHEMOTHERAPY USING IMU SENSOR DATA

Abstract

Gait analysis plays a crucial role in understanding motor impairments, an example of that is cancer patients undergoing chemotherapy who can be at risk of chemotherapy-induced peripheral neuropathy (CIPN). Inertial measurement units (IMUs) offer an objective and accessible means of tracking movement [1], enabling the extraction of gait features that may indicate early functional deterioration. By integrating signal processing techniques with machine learning (ML), this thesis aims to utilize IMU gait data to identify and segment gait events, assess its potential in biometric identification of different people under two walking conditions with the use of signal processing and machine learning techniques, and quantify gait changes over multiple visits for normal walking speed, all in cancer patients undergoing chemotherapy, providing a data-driven approach to assessing their mobility impairments.The first stage of this work focuses on preprocessing IMU data collected from 34 individuals under two different walking speeds, using signals obtained from foot-mounted IMUs, demonstrating that each person has their own unique walking pattern from features extracted from singular value decomposition (SVD) and discrete wavelet transform (DWT) used to train a support vector machine (SVM) classifier. This classification framework establishes the foundation for understanding individualized gait patterns and their variability. Building upon this, the later chapter shifts toward analyzing how gait evolves over time for each study subject undergoing treatment. The previously extracted (DWT) gait features are further examined using machine learning techniques for the purpose of identifying meaningful changes across multiple visits. These methods aim to track longitudinal changes and determine whether observed differences reflect inherent individual fluctuations or potential early indicators of functional decline. By integrating step-level analysis and feature-based modeling, this work seeks to provide a more comprehensive understanding of gait deterioration before it becomes clinically apparent, to enable earlier intervention, potentially improving the patient’s quality of life by addressing mobility issues before they impact a person’s daily activities.