Lumbar Disc Herniation Detection Based on Gait Analysis

Gait analysis is a simple and effective auxiliary diagnosis method for spinal diseases, and has achieved good results in the detection of Parkinson's symptoms and other medical diagnoses. Recently, gait-based assisted diagnosis of spinal diseases has received extensive attention in the research domain. Therefore, this paper studies the detection method of lumbar disc herniation based on patient gait data for auxiliary diagnosis of lumbar disc herniation. In order to obtain accurate gait features from video data efficiently, the human pose estimation tool OpenPose is used to preprocess the gait video data to obtain human key point sequence data. The gait features are quantized to form the vector representation of gait. According to the temporal characteristics of gait features, a TGAU model based on Transformer and gated attention unit(GAU) is proposed, where the GAU is used to effectively extract the features of gait data and reduce the influence of noise. Finally, we propose a model pre-training method based on transfer learning and use the open source OUMVLP-Pose gait recognition dataset to pretrain and fine-tune the model to enhance the performance. In order to verify the effectiveness of TGAU, we collect the gait video data of 10 patients with lumbar disc herniation and 11 normal people, obtain the gait analysis data through the proposed feature method. Experimental results show that our TGAU model gains higher accuracy than baselines, which demonstrate its effectiveness.