A complex network based approach for knee Osteoarthritis detection: Data from the Osteoarthritis initiative. (January 2022)

• Record Type: Journal Article
• Authors: Ribas, Lucas C.; Riad, Rabia; Jennane, Rachid; Bruno, Odemir M.
• Is Part Of: Biomedical signal processing and control. Volume 71(2022)Part A
• Publication Date: 2022-01
• Subjects: Texture analysis -- Networks -- Feature extraction -- OsteoArthritis detection; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.103133
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Complex network (CN) theory is used to model and extract image features. Osteoarthritis (OA) detection based on CN features extracted from Knee X-ray images. Experiments on OA initiative database validate its effectiveness. Results demonstrate that our approach outperforms the state-of-the-art studies. Abstract: OsteoArthritis (OA) is a joint disease caused by cartilage loss in the joint and bone changes. Early knee OA prediction based on bone texture analysis is a difficult task in medical image analysis. This paper presents a new approach based on concepts of complex network theory to extract texture features related to OA from radiographic knee X-ray images. An X-ray image is modeled into a complex network mapping each pixel into a node and connecting two nodes based on a given Euclidean distance. Then, a set of thresholds is applied to remove some edges and reveal texture properties. Our proposed model employs a specific strategy to automatically select the set of thresholds. A new set of statistical measures extracted from the network are used to compute a feature vector evaluated in a classification experiment using knee X-ray images from the OsteoArthritis Initiative (OAI) database. Our proposed approach is compared to state-of-the-art learning models (AlexNet, VGG, GoogleNet, InceptionV3, ResNet, DenseNet and EfficientNet) as well as to different traditional texture descriptors. Results show that the proposed method is competitive and is potentially promising for early knee OA detection.