Scholarlink

Neuropathy (NP) disease in lower limb is growing rapidly. Diabetic Sensorimotor Polyneuropathy (DSPN) a type of NP is a complication caused by Diabetic Mellitus (DM), currently affecting sufficient DM patients worldwide. It can create pain, diabetic foot ulceration (DFU), limb amputation, and even mortality. Early detection is crucial in identifying individuals at risk and it plays a pivotal role in preventing the DFU development and need for amputation. This paper presents a hybrid approach utilizing Deep Learning (DL) and Machine Learning (ML) for early stage detection of NP by using EMG data of patients. Pre-trained CNNs ResNet18, MobileNetV2, and SqueezeNet are hybridized with 3 ML models, K-Nearest Neighbor (KNN), Decision Tree (DT), and Support Vector Machine (SVM) classifiers and EMG spectrograms are used for the detection of NP. Results of proposed study show that SqueezeNet with SVM classifier perform better than other model with an accuracy of 85%, specificity of 86.90%, and F1-score of 82.35%. The best performing model is also hardware tested over Xilinx’s FPGA ZCU102 which provides accuracy with 10% tolerance which is in the software model. This study provides the foundation for the creation of edge devices for detecting NP and DSPN in future which will reduce the cost and time and also help the clinicians in detection of NP.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Performance Evaluation and Implementation of Hybrid Deep Convolutional Neural Network Models on FPGA for Biosignal Classification

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Sandeep Kumar Pandey,
  • Geetika Srivastava

摘要

Neuropathy (NP) disease in lower limb is growing rapidly. Diabetic Sensorimotor Polyneuropathy (DSPN) a type of NP is a complication caused by Diabetic Mellitus (DM), currently affecting sufficient DM patients worldwide. It can create pain, diabetic foot ulceration (DFU), limb amputation, and even mortality. Early detection is crucial in identifying individuals at risk and it plays a pivotal role in preventing the DFU development and need for amputation. This paper presents a hybrid approach utilizing Deep Learning (DL) and Machine Learning (ML) for early stage detection of NP by using EMG data of patients. Pre-trained CNNs ResNet18, MobileNetV2, and SqueezeNet are hybridized with 3 ML models, K-Nearest Neighbor (KNN), Decision Tree (DT), and Support Vector Machine (SVM) classifiers and EMG spectrograms are used for the detection of NP. Results of proposed study show that SqueezeNet with SVM classifier perform better than other model with an accuracy of 85%, specificity of 86.90%, and F1-score of 82.35%. The best performing model is also hardware tested over Xilinx’s FPGA ZCU102 which provides accuracy with 10% tolerance which is in the software model. This study provides the foundation for the creation of edge devices for detecting NP and DSPN in future which will reduce the cost and time and also help the clinicians in detection of NP.