Detection of Spinal Fracture Lesions Based on Improved Yolo-tiny

Spinal CT images are very important for the diagnosis of spinal diseases. However, due to the complication of spine CT images, and also artificial bias and low efficiency in clinical, which makes doctors to take many time in reading spine CT images, so it can not satisfy the clinical real time demands. In this paper, we prestent a method to detect spinal fracture lesions of (cervical fracture, cfracture), (thoracic fracture, tfracture), (lumbar fracture, lfracture) by proposed Yolo-tiny. We add three 1×1 convolution layers to the original Yolo-tiny network to make spinal lesions easier to detect. Our experimental results shows the mAP (mean Average Precision) of our proposed method is 86.63%, detection time is 0.02155 seconds per detection, IOU ((Intersection Over Union) is 87.3.