Real-time neurodegenerative disease video classification with severity prediction

Abstract

In this paper, an automatic diagnosis system for neurodegenerative diseases is presented. Starting with an existing neurodegenerative diseases gait dataset, namely the NDDGD dataset, classification and regression algorithms have been trained, with the inter-patient dataset separation scheme (walking patterns used for training and testing, belong to different people), and integrated within a larger automatic diagnosis system which make use of videos in input or real-time streaming from cameras for predicting the neurodegenerative disease, if present, and its stage. The proposed system is capable of predicting among 3 neurodegenerative diseases, namely: amyotrophic lateral sclerosis disease (ALS), Parkinson’s disease (PD), Huntington’s disease (HUN) and differentiate among the severity (stage) level of the disease, if found. The system makes use of common cameras for the 2D pose estimation and features engineering. The system can be easily deployed in hospitals and houses in order to help physicians with the diagnosis. When used in conjunction with physicians, this system can be a valuable tool for neurodegenerative diseases prediction.