Automatic Screening And Detection Of Real-time Diabetic Foot Wagner Grade Based On Deep Learning

As one of the most common complications of diabetes,diabetic foot has become one of the most difficult skin diseases due to its long treatment period and high cost.One of the effective methods for curing patients with diabetic foot is to reasonably classify and screen diabetic foot.In most countries and regions,the severity of diabetic foot patients still depends on professional podiatrists.This treatment method has the disadvantages of high treatment costs and longer treatment time.And in most underdeveloped countries and regions,due to lack of medical care facilities and inadequate awareness of patients,some patients with diabetic foot eventually become amputated as their illness worsens.It can be seen that the development of a remote,cost-effective and efficient automatic screening system for diabetic foot has become an effective method for diabetic foot diagnosis and patient self-management.In recent years,with the development of Computing power and deep learning,computer vision has made rapid progress,especially in the field of medical image analysis,such as nuclear magnetic resonance imaging,dermoscopy and diabetic foot.This paper proposes a method for automatic screening and detection of the Wagner grade of the diabetic foot based on deep learning.In this paper,we collected 2688 diabetic foot pictures as a dataset for deep learning model training and then used object detection algorithms such as Faster R-CNN,SSD,YOLOv3 to achieve automatic screening and detection of diabetic foot pictures.In this paper,the methods of image mixup,label smoothing,learning mode transition,and changing the number and size of Anchor Box are used to improve the above three types of object detection algorithms,and a stacked method is used to demonstrate the contribution of each method to the final accuracy.The experiments show that the mean average precision(mAP)of the refinements YOLOv3 algorithm reaches 92.0%,and the inference speed reaches 32 frames per second(FPS)under the acceleration calculation of the Tesla V100 GPU graphics card,which achieved the trade-off between speed and accuracy.In order to facilitate clinical use,this paper deploys the trained model to the Android mobile phone to achieve real-time diabetic foot Wagner grade detection.The experimental results show that the model can still achieve good real-time detection performance without the GPU graphics card,which provides an effective assessment for the analysis and healing status of diabetic foot ulcers.In summary,this paper provides new treatment ideas for the management of patients with diabetic foot and has higher clinical application value.