Correlative Postural Gait Descriptor to Discriminate Parkinsonian Findings in a Markerless Analysis

Parkinson's disease (PD) is one of the most prevalent neurodegenerative disorders worldwide, with over 10 million cases reported globally. Currently, gait analysis is crucial for quantifying motor abnormalities, and most methods rely on physical markers, which alter patients' natural movements and have limitations in explaining spatiotemporal relationships of the joints. This study aimed to implement a markerless methodology that reconstructs gait postures and encodes spatiotemporal joint information as covariance descriptors, which are then used to discriminate between subjects with Parkinson's disease and healthy controls. The methodology consisted of a study of covariance encoding of postures and joint trajectories to identify potential coordination patterns that aid in PD classification. The results obtained in a population of 30 subjects, recorded in a total of 240 videos, approved by the Scientific Research Ethics Committee of Universidad Industrial de Santander (CEINCI), showed an average PD classification accuracy of 75% and an AUC of 74%. In conclusion, the correlation descriptors demonstrated the potential to discriminate between subjects with PD and healthy controls and to encode coordination patterns. Furthermore, the visual correlation maps show differences between the two groups, which can further support routine clinical analysis.