JMIR Publications

ABSTRACT

Background:

Knee cartilage injury poses significant challenges in the early clinical diagnosis process, primarily due to its high incidence, the complexity of healing, and the limited sensitivity of initial imaging modalities.

Objective:

This study aims to employ Tomography and machine learning methods to enhance the classification accuracy of the classifier for knee cartilage injury, improve the existing network structure, and demonstrate important clinical application value.

Methods:

The proposed method-ology is a multi-dimensional feature Cross-Level fusion classification network driven by the Large Separable Kernel Attention, which enables high-precision hierarchical diagnosis of knee cartilage injury through deep learning. The network first fuses shal-low high-resolution features with deep semantic features via the Cross-Level Feature Fusion Module. Then, the Large Separable Kernel Attention module is embedded in the YOLOv8 network. This network utilizes the combined optimization of depth-separable and point-by-point convolutions to enhance features at multiple scales , thereby dramatically improving the hierarchical characterization of cartilage damage. Finally, five classifications of knee cartilage injuries are performed by clas-sifiers.

Results:

To overcome the limitations of traditional single-plane images, this paper constructs the first multidimensional MRI knee cartilage injury hospital real dataset, on which the classification accuracy is 99.7%, the Kappa statistic is 99.6%, the F-measure is 99.7%, the sensitivity is 99.7%, and the specificity is 99.9%. The experimental results validate the feasibility of the proposed method.

Conclusions:

In the context of the increasingly urgent demand for accurate diagnosis of medical imaging, this paper focuses on KCI, and proposes an MDF-CLF network based on the guidance of the LSKA to realize multi-classification diagnosis of knee cartilage injury through deep learning. By integrating sagittal, coronal, and transverse MRI images to construct a five-classification dataset of KCI, the algorithm proposed in this paper finally achieves an accuracy of 99.7%. Specifically, the study innovatively optimizes the YOLOv8 network backbone and introduces the LSKA kernel attention mechanism, which significantly enhances the model's ability to capture key features of the multimodal images, so that it can accurately focus on the injury region, identify subtle lesion features, and significantly improve the classification accuracy. The algorithm has been validated on real hospital datasets and performed well in the task of KCI classification, providing reliable technical support for clinical diagnosis. Although the algorithm has achieved stage-by-stage results, there is still room for optimization. Future research will focus on further improving the accuracy of the algorithm while expanding its application boundaries. In view of the fact that the medical imaging diagnosis of eye, lung, kidney, and other diseases also faces the problem of accurate classification, and there are commonalities in image characterization with KCI, the algorithm will be explored to be applied to the field of multi-disease classification, providing innovative solutions for more medical diagnostic scenarios, and helping medical imaging diagnostic technology to a new level.