A Computerized Meniscus Injury Detection System by using Convolutional Neural Network

Abstract

The meniscus provides a shock absorber between the shinbone and the thighbone. It is a C-shaped piece of tough, rubbery cartilage. One of the most frequent injuries to the knee is a torn meniscus. Meniscal tears can be confirmed clinically by magnetic resonance imaging, which can also review intra- and extra-articular anatomical structures and rule out other diagnoses. Techniques for image processing are now frequently used in many different applications. Meniscus extraction, image segmentation, and image enhancement are used in the image processing technique for meniscus tear detection. Convolution Neural Network (CNN) able to perform canny edge detection is regarded as superior because it can detect all existing edges in an image and is unaffected by noise. It also can identify thin edges in noisy images. Because the CNN method is so proficient at detecting lines in the sample image, it is used in this paper to detect meniscus tears. Based on elements like the quality of training data, model architecture, hyperparameters, and implementation details, a CNN model for meniscus tear detection may differ. To measure the efficacy of such models, evaluation metrics like sensitivity, specificity, precision, and F1 score are frequently employed. The primary goal of this work is to use MRI images to identify meniscus damage. On the provided sample images, the algorithm is tested. The model architecture and the level of expertise of the training data determine how accurate the CNN and the specific accuracy achieved for meniscus tear detection may vary depending on the specific dataset and problem domain, the accuracy of this model is typically 80% above. More accurate predictions imply more trustworthy outcomes.