Abstract— To cope with the high intra-subject variability of muscle activation intervals, a large amount of gait cycles is necessary to clearly document physiological or pathological muscle activity patterns during human locomotion. The Clustering for Identification of Muscle Activation Pattern (CIMAP) algorithm has been proposed to help clinicians obtaining a synthetic and clear description of normal and pathological muscle functions in human walking. Moreover, this algorithm allows the extraction of Principal Activations (PAs), defined as those muscle activations that are strictly necessary to perform a specific task and occur in every gait cycle. This contribution aims at assessing the impact of the number of gait cycles on the extraction of the PAs. Results demonstrated no statistically significant differences between PAs extracted considering different numbers of gait cycles, revealing, on average, similarity values higher than 0.88. Clinical Relevance—This contribution demonstrates the potential applicability of the CIMAP algorithm to the analysis of subjects affected by neurological disorders, for whom the assessment of motor functions may be of the uttermost importance and only a reduced number of gait cycles can be acquired. I. INTRODUCTION The analysis of surface electromyographic (sEMG) signals is commonly used to quantitatively assess normal and pathological muscle functions in human walking. Moreover, the assessment of sEMG signals can be a valuable tool in the evaluation of locomotion pathologies and rehabilitation protocols [1]. However, the great stride-to-stride variability of sEMG signals collected during gait [2], [3], even in healthy subjects, may strongly reduce the interpretability and reliability of the results. More specifically, previous studies reported that a subject’s walk can be characterized by 4-5 different muscle activation modalities (i.e., number of muscle activation intervals occurring within the same gait cycle) [2], [4] and different patterns within the same modality. Hence, a specific muscle may be activated with a variable sequence of patterns during the walking task. To cope with the high intra- cycle variability of the sEMG signals and to increase the interpretability and reliability of the results, the CIMAP (Clustering for Identification of Muscle Activation Pattern) algorithm was recently developed [5], [6] and validated on several healthy and pathological populations [7]–[9]. The Gregorio Dotti, Marco Ghislieri, Samanta Rosati, Valentina Agostini, Marco Knaflitz, and Gabriella Balestra are with the Department of Electronics and Telecommunications, Politecnico di Torino, 10129, Turin, and also with PoliToBIOMed Lab, Politecnico di Torino, 10129, Turin, Italy CIMAP algorithm is based on a hierarchical clustering method that allows the extraction of the different patterns. Its application permits the characterization of cyclical movements and the extraction of the Principal Activations (PAs). From a biomechanical point of view, PAs can be defined as those muscle activations that are strictly necessary to perform a specific task [6]. This algorithm allows clustering together the gait cycles showing similar sEMG activation intervals. Each cluster is described by an element (called cluster’s prototype) defined as the median of all the muscle activation intervals belonging to the same cluster. Then, PAs are computed as the intersection of all the representative clusters’ prototypes. In the optimized version of the CIMAP algorithm, to select the representative clusters, a cutoff threshold (ℎ) on the number of gait cycles per modality is applied. Clusters with a number of gait cycles per modality higher than this threshold are considered as representative and, hence, used for PA extraction, while clusters with a number of gait cycles per modality lower than this threshold are considered as non- representative [5], [7] and do not contribute to PA extraction. Therefore, the definition of this cutoff threshold is essential. On the one hand, the value of this cutoff threshold may affect the PA extraction. On the other hand, a high value of ℎ may limit the applicability of this approach to datasets containing a reduced number of gait cycles (e.g., 30-s lasting protocols [10]). The aim of this contribution is to assess the robustness of the cutoff threshold and to assess the effect of the number of gait cycles on the principal activation extraction, enhancing the applicability of the CIMAP algorithm to shorter acquisition protocols and different cohorts (i.e., patients affected by musculoskeletal or neurological disorders). II. MATERIALS AND METHODS A. Sample Population and Data Acquisitions Gait data from 20 healthy subjects (11 males and 9 females, age: 65.4 ± 5.1 years, height: 1.69 ± 0.09 m, weight: 69.0 ± 12.2 kg) were retrospectively analyzed from our gait data warehouse (BIOLAB, Politecnico di Torino, Turin, Italy). None of the subjects had neurological or orthopedic pathologies that could affect gait performance. Gait data were recorded using a multichannel acquisition system for gait analysis (STEP32, Medical Technology, Italy) with a sampling frequency of 2 kHz. More specifically, sEMG, foot- (e-mail: gregorio.dotti@studenti.polito.it, marco.ghislieri@polito.it, samanta.rosati@polito.it, valentina.agostini@polito.it, marco.knaflitz@polito.it, and gabriella.balestra@polito.it) The Effect of Number of Gait Cycles on Principal Activation Extraction G. Dotti, M. Ghislieri, Student Member, IEEE, S. Rosati, Member, IEEE, V. Agostini, Member, IEEE, M. Knaflitz, Member, IEEE, and G. Balestra, Member, IEEE 2021 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) Oct 31 - Nov 4, 2021. Virtual Conference 978-1-7281-1178-0/21/$31.00 ©2021 IEEE 985