Hybrid adaptive approaches applied to gait analysis and therapeutic decision support

• Angewandte hybride adaptive Methoden zur Ganganalyse und Unterstützung therapeutischer Entscheidungen

Rêgo Caldas, Rafael; Markert, Bernd (Thesis advisor); Loosen, Peter (Thesis advisor); Buarque de Lima Neto, Fernando (Thesis advisor)

Aachen : RWTH Aachen University (2020, 2021)
Book, Dissertation / PhD Thesis

In: Report. IAM, Institute of General Mechanics 06
Page(s)/Article-Nr.: 1 Online-Ressource : Illustrationen, DiagrammeReport number: IAM-06

Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2020

Abstract

The walking pattern is frequently affected by several injuries on the lower limbs and also neurological disorders. Gait analysis is the systematic study of the walking pattern, providing information about the subject’s functional level. However, the conventional methods to assess the gait are either expensive or complex to be applied regularly in the clinical practice. In this context, age-related changes in walking parameters are relevant to analyze the functional capacity and physical performance. Moreover, the nuances of slightly different gait patterns are hardly noticeable by inexperienced observers. Given the complexity and importance of the gait pattern and its analysis in a clinical context, the main goal of this research work is to improve the conventional evaluation by enhancing the option set to establish efficient rehabilitation methods. To this end, two hybrid adaptive computational approaches and an application of predictive methods were proposed to group subjects according to gait-related features and to determine the likely rehabilitation duration, respectively. For the first contribution, kinematic features of healthy volunteers were utilized for subject clustering according to their age and the essential variations on their gait pattern. The experiments, which used the Self-organizing Maps (SOM) algorithm associated with Fuzzy c-Means (FCM) and k-Means (KM) as clustering methods, have presented insightful outcomes for the validation of these two hybrid adaptive approaches. The results pointed out compatible performances of the proposed approaches, overpowering the basic FCM algorithm. Since the SOM+KM requires less computational resources to cluster the subjects while delivering a similar performance, we recommend its application rather than the approach with FCM. Besides the more consistent results provided by our hybrid approaches, as a second contribution, we also proposed an analysis of feature relevance based on the SOM output as heat maps. This technique has the advantage of preserving the data topology in the produced graphic output. All analyses were carried out in two different contexts of volunteers: (i) grouping subjects in age-related clusters based on gait features and (ii) grouping subjects regarding different speeds and placement of sensors. Among the kinematic temporal parameters, the results suggest the overall importance of cadence, as a measurement of physical performance, especially when clustering subjects by their age. Concerning the second relevance analysis, the component of individuality stood out while observing results of different volunteers. Furthermore, the proposed adaptive approaches have evoked encouraging results on the prediction of the rehabilitation duration, considering the clinical data based on the physical exam, therapeutic goals, adopted techniques, and the required number of physical therapy (PT) sessions. Two of the four methods tested have demonstrated high accuracy to determine the number of PT sessions necessary to recover gait-affecting injuries on the lower limbs. Such innovative applications can support decisions not only of physicians and patients but also healthcare-facilities managers, concerning the best alternatives for the patients and hospitals, impacting the clinical practice positively.