Multi-level classification of knee cartilage lesion in multimodal MRI based on deep learning. (May 2023)

• Record Type: Journal Article
• Authors: Zhang, Lirong; Che, Zhiwei; Li, Yang; Mu, Meng; Gang, Jialin; Xiao, Yao; Yao, Yibo
• Is Part Of: Biomedical signal processing and control. Volume 83(2023)
• Publication Date: 2023-05
• Subjects: 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.2023.104687
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Extraction of multimodal KC features based on dragonfly optimization and regional similarity transformation algorithm (DO-RSTA). It significantly enhances the accurate information of the edges and areas of the KC, avoids the loss of sensitivity caused by the MRI volume effect, and obtains the accurate feature of the edges of the KC and the adjacent confusing areas. The proposed algorithm reconstructed the deep learning model, which can better integrate the multimodal features into the global multi-scale features, then improve the feature fitting ability and training efficiency of KC lesion images. The proposed model is trained with the actual hospital medical image dataset to minimize medical cost and patient pressure, which can achieve the accurate and efficient automatic classification of KC injury employing MRI. Abstract: The regeneration and repair ability of knee cartilage is limited, and the early clinical symptoms of patients are not obvious, so the diagnosis of knee cartilage damage is crucial for clinical treatment. To effectively overcome the irreversible injury caused by minimally invasive arthroscopic knee surgery, a multi-classification model of knee cartilage injury based on deep learning is proposed. The model has the characteristic of multi-feature fusion, which can realize automatic non-invasive monitoring and examination of knee cartilage injury.Furthermore, the proposed algorithm uses dragonfly optimization and regional similarity transformation algorithms to extract valid regional information on knee cartilage, cartilage edema, and subchondral bone in different modalities and integrates it into global multiscale features. It can obtain accurate information on the edges of knee cartilage and adjacent confusing areas, and solve the problem of less authentic medical images in hospitals for data enhancement, to realize an accurate network model of knee cartilage injury classification. The proposed algorithm performs a five-level classification of authentic hospital data sets with an accuracy of 99.73%. The experimental results show that the proposed model is generally higher than the current state-of-the-art classification depth model. Keywords: Knee cartilage lesion; Multilevel classification; Deep learning; Multimodal features.