A vision‐based clinical analysis for classification of knee osteoarthritis, Parkinson's disease and normal gait with severity based on k‐nearest neighbour

Abstract

The prevalence of musculoskeletal and neurological diseases such as knee osteoarthritis (KOA) and Parkinson's disease (PD) has grown speedily in recent years. The direct impact of these disorders on a person's gait has evoked many researchers to perform their analysis using gait because of its greater feasibility. In this paper, we proposed a framework to accurately classify abnormal considering KOA and PD and normal (NM) gait using a vision‐based (VB) approach. The aim of this study is four‐fold: firstly, a novel VB gait dataset is created and presented; secondly, the segmentation of the region of interest (ROIs) is performed using an improved technique; thirdly gait parameters, that is, spatiotemporal (SPT), linear kinematic (KNM) and additional body motion features are evaluated using statistical methods; fourthly, the classification of KOA, PD and NM subjects has been done based on artificial intelligence or a machine learning technique (MLT), that is, k‐nearest neighbour (KNN) at three severity levels and a comparison is performed with other methods namely support vector machine (SVM), random forest (RF) and linear regression (LR). The results reveal the relevance of considered gait features in this investigation. For classification purposes, the highest results are obtained by our proposed method in classifying the subjects with severity also, achieving an overall accuracy = 0.9006, sensitivity = 0.8554, specificity = 0.9044, and precision = 0.8966 respectively. This study thus successfully presents KOA, PD, and NM gait classification method based on the MLT which may be beneficial for clinicians to perform early diagnosis of such disorders objectively.

Improving Knee Osteoarthritis Classification with Markerless Pose Estimation and STGCN Model

“…Furthermore, one of the notable features of BlazePose is its real-time performance: it runs at high frame rates, allowing for instant tracking and analysis of all 33 joints. This augmentation in the number of articulations significantly improves upon the previous work, which only utilized six joints [3]. Fig.…”

“…In this study, we utilized the advanced technique of BlazePose for video analysis. BlazePose is a markerless joint detection method that predicts (x , y , z ¿ coordinates for each joint from weakly controlled videos, unlike traditional methods that require placing markers on subjects [3], [10]. One of the remarkable features of BlazePose is its ability to detect and track 33 joints of the human body.…”

“…Furthermore, the use of marker-based approaches aids in the accurate detection of keypoints, enabling effective feature extraction. In [3] the authors introduce a framework that effectively classifies abnormal gait patterns associated with KOA and Parkinson's disease (PD) from normal (NM) gait. They employ a vision-based (VB) approach, leveraging a newly created and presented Vision-based Gait Dataset.…”

Exploration and demonstration of explainable machine learning models in prosthetic rehabilitation-based gait analysis

“…Other ML methods, including support vector machine (SVM) algorithms, were successfully used for the automatic classification of prosthetic components, despite relatively small sample sizes [ 10 ]. In [ 11 ], a k-nearest neighbour method was successfully used to classify the abnormal gait patterns caused by knee osteoarthritis and Parkinson’s disease. In [ 12 ], four different ML models, including partial least square-discriminant analysis, SVM, random forest (RF), and artificial neural networks, were used to extract and use informative features for classifying gait characteristics according to certain subject characteristics.…”

“…Recently, there has been a growing interest in the application of ML models for gait analysis methods, as they can address some of the shortcomings of traditional statistical methods and provide flexibility, scalability, and independence from inferred assumptions [10][11][12][13][14]. ML methods can also be applied to different data types (e.g.…”