Machine Learning-Based Self-Induced Scratch Intensity Detection Using Feature Optimization and Multi-Channel Electromyogram Signals for Prevention of Lichenification.

Chronic pruritus in patients with dermatological conditions causes physical discomfort, skin breakdown, sleep disturbance, and overall decline in quality of life. Persistent scratching can lead to a condition known as lichenification, which further deteriorates the skin. Lichenified skin is more susceptible to infections, which further complicates the treatment and prolongs the recovery time. This highlights the importance of accurately detecting and quantifying scratching behavior for effective management and intervention. All the existing technologies to monitor scratching involve the use of external sensor modalities such as an accelerometer and a gyroscope, followed by camera monitoring. However, such scratch detection methods are limited in their reliability in monitoring scratching intensity. This study aims to acquire muscle activity to accurately detect, quantify, and provide feedback on self-induced scratching intensity. Through a meticulous understanding of the synergies of hand muscles, three out of seven forearm muscles were identified that generate consistent and distinctive signals during scratching motions. These signals were acquired, preprocessed, segmented, and analyzed using both time- and frequency-domain features. The extracted 45 features of 3-Channel EMG signals were first optimized using recursive feature elimination and validated by cross-validation accuracies of the recursive feature elimination technique, with the highest optimized accuracy of 0.8628 attained when EMG of a complete combination of muscles is used. This study shows promising results in facilitating enhanced diagnosis and management of scratching intensities. The model may support future wearable devices for generating alerts and providing advanced therapy.