Intelligent detection of knee injury in MRI exam

Knee injuries are one of the most common injuries that occur, especially among athletes and older people. They are broadly classified into three main kinds of injuries—meniscal tear, anterior cruciate ligament (ACL) tear and abnormality. The best and most preferred method for determining knee injuries' severity is magnetic resonance imaging (MRI). Despite this, knee MRI interpretation is time-consuming and subject to diagnostic error and variability, resulting in many unnecessary surgeries and false-positive predictions. As a result, developing an automated system to interpret knee MRI could help clinicians prioritize patients at a higher risk and make better, more accurate diagnoses. This can be achieved with the help of deep learning methods, which should be capable of automatically learning the layers of features and must be capable of modelling the dynamic relationships between medical images and their interpretations. This paper aims to solve and go through the problem of Knee injuries detection in medical diagnosis and solve the problem by processing MRI by building a multi-model convolutional neural network (CNN) consisting of four pre-trained models—VGG16, VGG19, ResNet152V2, InceptionV3, DenseNet201 to help classify knee injuries from MRI scans into ACL tears, meniscal tears or abnormalities in the knee. The proposed model shows the average highest accuracy, 78.33% using ResNet152V2 as compared with state-of-the-art work for three-class classification of Knee injuries using three different planes of MRI scan.