Comparative study of object detection models for abnormality detection on spinal X-ray images

Detecting abnormalities on spinal X-ray images plays a crucial role in early diagnosis and treatment of spinal disorders. In this paper, we propose a comparative study of three models namely Faster R-CNN, RetinaNet, and FCOS for abnormality detection on spinal X-ray images. Experimental results were measured by the mean average precision at a threshold of 0.5 (mAP@0.5). We investigate the impact of varying input image sizes and the number of updated weight layers in ResNet FPN backbone for Faster R-CNN model. Among the models, the Faster R-CNN model with the ResNet152 FPN backbone achieves the highest accuracy. Additionally, these analysis and evaluation results can serve as reference material for researchers working in the field of abnormality detection on model selection and configuration aiming to enhance diagnostic outcomes.