Deep-learning models show potential for diagnosing neurodegenerative diseases

Early diagnosis of neurodegenerative diseases is challenging, and there is a need for new methods.
Two deep-learning models have been developed with the aim of classifying amyotrophic lateral sclerosis, Parkinson’s disease, and Huntington’s disease from each other and from healthy controls, using gait signals transformed onto spectrogram images.
The first model – in which spectrogram images are fed directly into a convolutional neural network–long short-term memory (LSTM) network – had 99.42% accuracy for distinguishing the diseases, while the second – in which wavelet transform was used before the LSTM unit – had accuracy ranging from 94.04% to 95.37% depending on how many sub-bands were used.