Meniscus tears recognition by deep neural networks ; Розпізнавання пошкоджень меніска засобами глибинних нейронних мереж

In the realm of contemporary medical research, a critical task is enhancing the accuracy of diagnostics and treatment for injuries to the knee joint, especially the meniscus. Given the limitations of traditional methods like arthroscopy, which often rely on the surgeons experience and skills, there is an emerging need for developing and applying innovative technologies. Proposed to use deep neural networks (DNNs) for automated recognition of meniscal tears during surgical operations based on images from arthroscopic cameras. Various configurations of YOLOv8 architecture, ranging from Nano to Large, are considered, enabling the tailoring of solutions to specific clinical needs and computational resources. While deep learning in medicine is not a novel phenomenon, this work emphasizes the innovation in adapting powerful algorithms to the specific challenges associated with diagnosing the condition of intra-articular structures. A dataset was created to develop an effective diagnostic system that includes annotated images of meniscal tears. This dataset was utilized to train and test models, allowing for the analysis and comparison of each models ability to generalize and accurately recognize pathologies. Previous works that explore the application of classical deep learning methods to medical diagnostic systems were reviewed. The studys findings revealed that the application of DNNs significantly enhances medical images accuracy and processing speed. However, considering the complexity and diversity of meniscal injuries, it was decided to use deep learning models for detailed analysis of large datasets. During the research, models based on deep learning were trained, which significantly increased the accuracy of identifying types of damage. Experimental tests conducted based on the data obtained showed that models trained on expanded datasets demonstrated higher classification accuracy. These results confirm the potential of using deep learning to improve medical diagnostic capabilities. The research could make a ...