The Study Of Computer-aided Diagnosis Method For Knee Osteoarthritis Based On Deep Network

Knee Osteoarthritis(KOA)is one of the most common joint diseases.At present,the main technologies for diagnosis of KOA are imaging examinations,including X-ray,MRI,etc..However,these technologies are expensive and not convenient in everyday life.Compared with these methods,the vibroarthrography is a new diagnostic technology for KOA that is being explored in clinical practice in recent years,which can be obtained cheaper and more conveniently.But it is challenging for doctors to evaluate the patients’ condition by visually detecting VAG signals due to the limited understanding about information contained in VAGs.Hence,there has been an increasing interest in computer-assisted diagnosis of KOA(KOA-CAD)using VAG signals with the help of machine learning approaches in recent years.This paper focuses on two tasks of KOA-CAD(the early screening of KOA and the grading detection of KOA),and studies the computer-assisted diagnostic method of KOA using VAG signals.In this paper,an early KOA screening method based on convolutional network using VAG signals is first proposed.Here,the Fourier transformation method is applied to calculate the power spectrum and phase spectrum of VAG signals.And then a frequency domain-based convolutional neural network(F-CNN)is designed.At last,F-CNN model is trained to complete automatic identification between normal subjects and mild KOA patients.Besides,a KOA grading detection method based on ensemble network using VAG signals and physiological information is proposed.Here,a CNN block(VAG-CNN-Block)is first constructed by adjusting the F-CNN model structure,which aims to analyze the VAG signals.Then,a feedforward neural network block(PI-FNN-Block)is constructed which aims to analyze the physiological information.Furthermore,a model-based ensemble Network(MBE-Net)is designed by attention mechanism.At last,the MBE-Net model is trained to realize automatic recognition of normal subjects,mild KOA patients and severe KOA patients.Finally,the early screening method and the grading detection method are verified by VAG signals and physiological information,which are collected from two hospitals in Xi’an.The early screening accuracy,sensitivity and specificity attain 86.2%,88.2% and 83.3%respectively.The grading detection accuracy,sensitivity and specificity attain 87.5%,87.2%and 93.6% respectively.