Revisiting the K-Fold Approach for a Stable Model on Amyotrophic Lateral Sclerosis Prediction Scheme using LSTM and Attention Mechanism

A progressive neurodegenerative disease affecting motor neurons, Amyotrophic Lateral Sclerosis (ALS) requires early diagnosis as quickly as possible. For such situations, surface electromyography (S-EMG) is widely used as a non-invasive diagnostic tool to measure muscles' activity through electrodes placed on the skin's surface. Artificial intelligence (AI) approaches can be employed to analyze captured signals and distinguish abnormal patterns. However, previous work focused primarily on spatial information. It does not consider temporal information, effectively capturing the dynamic nature of muscle activity and identifying subtle abnormalities that might indicate ALS. Therefore, we fill the gap in this study by proposing a combination of CNN, Long-Short-Term Memory Networks (LSTM), and attention mechanisms to exploit temporal information in EMG signals. Stability assessment using K-fold cross-validation ensures reliable model performance. Our results demonstrate that combining spatial and temporal information can enhance performance and acquire 98.15% and 98.45% for CNN and LSTM, and CNN, LSTM, and Attention combination. In addition, our proposed model remains stable compared to previous work.