Machine Learning for Prosthesis Motion Intent Classification Using Multiple Non-invasive Sensing Modalities

The field of prosthetics has seen significant advancements in recent years, offering artificial devices to replace or augment lost limbs. However, high abandonment rates of currently available prosthetics persist due to lack of intuition and sensory feedback. This research aims to address these limitations by proposing a novel technique to improve the intuition of smart prosthetics. The main objective is to develop a robust and accurate classifier that detects and classifies the user's hand motion intent, enhancing mobility and functionality for individuals with limb loss. The proposed prototype integrates non-invasive techniques, including Force Mechanomyography (FMG), Inertial Measurement Unit (IMU), and pneumatic mechanomyography (pMMG). A dataset for 8 hand gestures is obtained from 10 test subjects, and a classification system using Support Vector Machine (SVM) and Linear Discriminant Analysis (LDA) Dimensionality Reduction is developed and tested against 3 transradial amputee subjects. The study focused on 8 hand gestures in this research, including extension, flexion, pronation, supination, radial deviation, ulnar deviation, hand closing, and hand opening—representing basic hand movements. The proposed configuration offers a potential breakthrough in prosthetic control, as it leverages multi-modal sensor inputs and machine learning techniques to enhance user experience by identifying patterns and adapting to variations in data. This technological advancement holds promise for medical professionals to provide more effective and customized solutions to patients.