Detection of Anterior Cruciate Ligament and Meniscal Tears on Knee Mr Images Using Machine Learning and Deep Learning

Abstract

Anterior Cruciate Ligament (ACL) tears are often sustained by football players, volleyball newlineplayers, sprinters, runners, and other athletes. This typically occurs due to excessive stretching newlineor sudden, abrupt movements, causing intense pain for the individual. While numerous computer newlinevision-based techniques have been employed to detect ACL tears, the complex structure newlineof knee ligaments presents significant challenges to the performance of most systems.A novel newlinemultidirectional multi-neighbor local binary pattern (MNLBP) texture descriptor is presented newlinefor the detection of ACL normal and tear MR image of knee ligament in the initial phase of the newlinestudy. KNN and SVM classifiers are used to assess the performance of MNLBP. The MRNet newlineknee MR image dataset is used for training as well as testing the system. The performance newlinemeasure of this methodology is achieved with the help of sensitivity, specificity, accuracy, and newlinethe F1 score. The suggested MNLBP performed better than conventional LBP for the KNN newlineand SVM classifiers, achieving a specificity of 0.92 and 0.92, a sensitivity of 0.87 and 0.88 and newlinea precision of 88. 92% and 89. 41%. The second phase of research proposed an effective and newlinesimple approach for detecting ACL tears using a convolutional neural network (ATD-CNN) newlineand Meniscal tear to minimize the complexity of the network. Data augmentation based on newlineshifting and rotation is used to develop the synthetic database to diminish the data scarcity issue. newlineFor the original and supplemented datasets, ATD-CNN yields an accuracy of 90. 10% and newline93. 93%, respectively. The third research phase is to improve the feature uniqueness of knee newlineMR images for the detection of Anterior Cruciate Ligament tears and meniscal tears using a newlinethree-layered MultiKernNet DCNN (MultiKernNet). Performance of the proposed methods newlineare assessed with Precision, Recall, Accuracy, and F1-score. Overall Accuracy of the Multi- newlineKernNet is of 96.60%, precision of 0.9654, recall rate of 0.9668, and F1-score of 0.9582. The newlineproposed ATD-CNN and MultiKernNet