Postural regulation and signal segmentation using clustering with TV regularization approach

• Published in: Biomedical signal processing and control 2025-01, Vol.99, p.106808, Article 106808
• Main Authors: Trabelsi, Imen, Hérault, Romain, Baillet, Héloise, Thouvarecq, Régis, Seifert, Ludovic, Gasso, Gilles
• Format: Article
• Language: English
• Subjects: ADMM; Clustering; Computer Science; Postural coordination; Total variation
• ISSN: 1746-8094
• DOI: 10.1016/j.bspc.2024.106808

This paper investigates a clustering algorithm with Total Variation (TV) constraint for postural regulation from postural coordination signals. The problem addressed aims to automatically segment postural coordination signals into behavioral patterns according to the patients’ performances. Starting from the assumption that the strategies of postural regulation by a patient will be almost constant during an experimental session and for the same oscillation frequency of the mechanical horse, we propose to add to the clustering problem a TV regularization which forces two consecutive cycles to belong to the same cluster. The resulting optimization problem is solved by an ADMM approach. An extensive experimental evaluation is presented, where we demonstrate the usefulness of our approach on various tasks. The proposed clustering scheme was tested on several real datasets, including non-temporal data, time-series signals, and action temporal segmentation data. Experimental findings show that the approach significantly increases the clustering accuracy with gains of 3.7% to 51.2% on non-temporal data and compares favorably with other methods on time series clustering. Also, improved clustering precision was reported on action segmentation task. Finally, our TV regularization approach was evaluated on postural data from brain-damaged people. Empirical results illustrate the meaningfulness of the obtained clusters and signal segmentation. •We propose a new clustering algorithm using temporal coherence and TV regularization.•We illustrate the algorithm’s effectiveness on time-series data and action segments.•We show that hippotherapy improves postural coordination in brain-damaged patients.