A Comprehensive Deep Learning Approach to Improve Enchondroma Detection on X-Ray Images

Abstract

An enchondroma is a benign neoplasm of mature hyaline cartilage that proliferates from the medullary cavity toward the cortical bone. This results in the formation of a significant endogenous mass within the medullary cavity. Although enchondromas are predominantly asymptomatic, they may exhibit various clinical manifestations contingent on the size of the lesion, its localization, and the characteristics observed on radiological imaging. This study aimed to identify and present cases of bone tissue enchondromas to field specialists as preliminary data. In this study, authentic X-ray radiographs of patients were obtained following ethical approval and subjected to preprocessing. The images were then annotated by orthopedic oncology specialists using advanced, state-of-the-art object detection algorithms trained with diverse architectural frameworks. All processes, from preprocessing to identifying pathological regions using object detection systems, underwent rigorous cross-validation and oversight by the research team. After performing various operations and procedural steps, including modifying deep learning architectures and optimizing hyperparameters, enchondroma formation in bone tissue was successfully identified. This achieved an average precision of 0.97 and an accuracy rate of 0.98, corroborated by medical professionals. A comprehensive study incorporating 1055 authentic patient data from multiple healthcare centers will be a pioneering investigation that introduces innovative approaches for delivering preliminary insights to specialists concerning bone radiography.