Portesan, Tommaso (2025) Decoding grasping-related muscle activations using deep and shallow machine learning models. Master thesis, Psychology.
|
Text
MASTER-THESISTommasoPortesanAIS-5.pdf Download (4MB) | Preview |
Abstract
Grasping is a critical motor function that involves complex neuromuscular interactions, with significant potential to advance fields such as prosthetics and rehabilitation. This master's thesis explores the decoding of grasping-related muscle activations through the application of deep and shallow machine learning models. The main objectives were to elucidate the muscular mechanisms underlying grasping and to assess the distinguishability of grasp types and object properties using electromyography (EMG) signals. EMG data were previously collected from 16 healthy participants performing tasks with three grasp types (power grasp, five-finger precision, two-finger precision) on four objects (large/small cylinders and spheres). The EMG analysis was based on five distinct phases of the grasping movement: fixation, observation, planning/execution, holding, and releasing. The methodology employed Representational Dissimilarity Matrices (RDMs) to evaluate the distinctiveness of muscle activation patterns, Linear Discriminant Analysis (LDA) for shallow classification, and Convolutional Neural Networks (CNNs) for deep learning-based classification. Key findings indicate that muscle activity patterns were most distinct during the holding, planning/execution, and releasing phases, with CNNs achieving up to 88.77% accuracy (chance level 33%) in classifying grasp types. However, object type classification was less successful (peak accuracy of 43.29%, chance level 25%), suggesting that EMG signals better reflect hand configurations than object characteristics. These results highlight the efficacy of machine learning in interpreting grasping dynamics and point to promising applications in developing advanced prosthetic systems, neuromotor rehabilitation, and brain-computer interfaces, offering new avenues to enhance motor control technologies.
| Item Type: | Thesis (Master) |
|---|---|
| Supervisor name: | Sburlea, A.I. |
| Degree programme: | Psychology |
| Differentiation route: | Other [Master Psychology] |
| Date Deposited: | 17 Sep 2025 10:09 |
| Last Modified: | 17 Sep 2025 10:09 |
| URI: | http://gmwpublic.studenttheses.ub.rug.nl/id/eprint/5932 |
Actions (login required)
 |
View Item |