Related Experiment Video
Updated: Aug 7, 2025
09:47
Author Spotlight: Advancing Alzheimer's Research – Exploring Early Detection and Multi-Omics Approaches
Published on: December 15, 2023
1.2K
Early Diagnosis of Neurodegenerative Diseases Using CNN-LSTM and Wavelet Transform
Elmira Amooei1, Arash Sharifi1, Mohammad Manthouri2
1Department of Computer Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran.
Journal of Healthcare Informatics Research
|March 13, 2023
Summary
This study introduces two deep learning models for early neurodegenerative disease diagnosis using gait analysis. One model achieved 99.42% accuracy, while a wavelet-enhanced model offers faster, lighter prognostics for diseases like Parkinson's.
Area of Science:
- Medical Technology
- Artificial Intelligence in Medicine
- Biomedical Signal Processing
Background:
- Early diagnosis of neurodegenerative diseases (NDs) like Amyotrophic Lateral Sclerosis (ALS), Parkinson's disease (PD), and Huntington's disease (HD) remains a significant clinical challenge.
- Deep neural networks (DNNs) show promise for rapid and precise disease diagnosis in healthcare.
- Gait signals offer a non-invasive method for assessing neurological function and detecting subtle changes associated with NDs.
Purpose of the Study:
- To develop and compare two novel Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) models for classifying ALS, PD, HD, and healthy controls.
- To investigate the efficacy of using gait spectrogram images as input for DNN-based classification of NDs.
- To evaluate the impact of wavelet transform as a feature extraction technique on the performance and efficiency of ND classification.
Main Methods:
- Two CNN-LSTM models were designed to classify NDs using gait spectrogram images.
- Model 1 directly processed spectrogram images with a CNN-LSTM network.
- Model 2 incorporated wavelet transform for feature extraction prior to the LSTM layer, with subsequent analysis of sub-band contributions.
Main Results:
- Model 1 achieved a high classification accuracy of 99.42%.
- Model 2, utilizing wavelet transform, demonstrated significant improvements in efficiency, requiring approximately 103 times fewer training parameters.
- Classification accuracy using Model 2 varied with sub-band usage: 95.37% with approximation sub-bands, 94.04% with three sub-bands, and 94.53% with all sub-bands.
Conclusions:
- Deep learning models, particularly CNN-LSTM architectures, are highly effective for classifying neurodegenerative diseases from gait spectrograms.
- Wavelet transform, especially using approximation sub-bands, can enhance model efficiency (lighter, faster) without substantially compromising diagnostic accuracy.
- The findings suggest a viable pathway for developing computationally efficient tools for early ND detection using gait analysis.
