Deep Learning based Dual Image and EMG Signal Low Back Pain Detection Methodology

Non-specific low back pain (nLBP) has steadily increased in prevalence with the rise of sedentary and inactive lifestyles among the general population. It is important to identify the affected muscles so that therapy may be tailored to the individual. Despite this, current approaches for recognizing affected muscles rely heavily on clinicians' expertise and lack objective criteria. Most biomedical signal-based diagnostic tools, including surface electromyography (sEMG), can only distinguish between patients and healthy controls. Since it relies on human-created characteristics, EasiSMR is only one work capable of identifying symptomatic muscles; nonetheless, its accuracy is limited. In this article, we offer Proposed work, a system that uses deep learning to identify the muscles that are causing non-specific low back pain. Its raw sEMG signal is converted into both frequency and time domain sEMG first. The data is then sent into a heterogeneous two-stream multitasking deep learning system that analyses each input independently based on its unique properties. In addition, we include the muscles' compensation information into our multitask neural network architecture and propose Spanning CNN to enhance recognition accuracy. Finally, we verify our system's performance by designing and implementing a waist-belt-shaped wirelessly sEMG monitoring and processing system.