Gait-based neurodegenerative disease classification using multimodal temporal-spectral representations and a dual-branch CNN.

Neurodegenerative diseases (NDDs) often cause gait impairments with overlapping motor symptoms, complicating multi-class discrimination. This study proposes a lightweight dual-branch convolutional neural network integrating recurrence plots and spectrogram representations of vertical ground reaction force signals to classify Parkinson's disease, amyotrophic lateral sclerosis, Huntington's disease, and healthy controls. Evaluated on the Gait in NDDs Dataset comprising 64 subjects, the proposed framework achieved 96.15% accuracy, outperforming recurrence-based (94.19%) and spectrogram-based (95.75%) single-branch models. The findings demonstrate that combining complementary temporal recurrence and spectral information improves discrimination of neurodegenerative gait patterns while maintaining low computational complexity.