Comprehensive Summary
This study focuses on using deep learning models to improve efficiency and accuracy of enchondroma detection in X-rays. Two convolutional neural network (CNN)-based object detection frameworks, Detectron2 and YOLOv8, were compared in terms of accuracy and processing time. These models were selected for their high mean average precision scores, quicker inference speeds, and recent technological advancements. 1,055 X-rays were compiled into a dataset, with 844 used for training and 211 for validation. Orthopedic specialists labeled the enchondromas on the images and each area of interest was then masked to increase performance and efficiency. The main evaluation methods used were mean Average Precision (mAP), accuracy, and recall. The mAP of YOLOv8 was 97.4%, and its overall detection rate was greater than 95%. However, Detectron2 performed slightly better than YOLOv8, with an accuracy score of 98.3%. These findings suggest that both models have potential to support clinical decision-making and diagnostics in orthopedics.
Outcomes and Implications
Deep learning image detection models such as YOLOv8 and Detectron2 can provide a support system for decision making and detection of enchondromas. Using an AI-based detection model for the detection of enchondromas can allow physicians to spend more time on the treatment of patients, reducing time between diagnosis and treatment. Additionally, these models can be used to continuously model enchondromas to ensure they do not progress or become malignant. Further research is needed to expand the use of these tools in tumors at different stages, as well as for progression and long-term monitoring.