ALSNet: A Lightweight Deep Learning Framework for ALS Detection Using Enhanced TF Images From EMG Sensor Data

Muscle electrical activity is evaluated using electromyography (EMG), which offers valuable information for identifying neuromuscular conditions like amyotrophic lateral sclerosis (ALS), a degenerative disorder affecting motor neurons. ALS is a neurode-generative condition, that requires accurate and swift diagnosis for effective management. This research presents a novel method for detecting ALS using enhanced time-frequency (TF) images generated from filtered EMG sensor data. In this research, EMG signals are converted into time-frequency (TF) images using wavelet synchrosqueezed transform (WSST) and continuous wavelet transform (CWT), capturing both time and frequency components of EMG signals, which are essential for extracting relevant features. To enhance the quality and contrast of these TF images, researchers applied an advanced technique known as contrast limited adaptive histogram equalization (CLAHE). To create an automated system for detecting ALS, the obtained enhanced TF images are input into benchmark models like Squeeze-Net, Alex-Net, and a newly designed, lightweight deep learning model. The proposed method exhibited improved accuracy and robustness compared to existing techniques. The deep learning model, utilizing the enhanced TF images, achieved a high accuracy of 94.89% and ROC of 0.99 in ALS detection, surpassing other current methods. This investigation emphasizes the importance of preprocessing and enhancement in diagnostic applications that combine TF analysis with deep learning approaches. The findings demonstrate the potential of this novel method for reliable ALS detection, paving the way for future advancements in clinical diagnostics.