Spinal Cord Injury Identification and Localization Detection Based on MRI Imaging and Deep Learning Technology

Abstract

This study explores the feasibility and application value of the Faster R-CNN algorithm, a deep learning technology, in identifying signal abnormalities and locating injury areas in MRI images of the spinal cord. Method: Initially, Magnetic Resonance Imaging (MRI) images from 1,000 spinal cord injury (SCI) patients and 500 healthy individuals collected over five years were included in the dataset, divided into signal change and normal groups. The dataset was then preprocessed, and the lesion areas were annotated by experienced spine surgeons for later experimental verification of the algorithm's effectiveness; no markings were necessary for the normal group. Subsequently, the Faster R-CNN algorithm, combined with the VGG-16 and Resnet50 network models from the convolutional neural network (CNN) framework, was used for recognizing and locating SCI in MRI images. Finally, a horizontal comparison of different network structure models was conducted, with the model's mean Average Precision (mAP) and visual results serving as evaluation metrics to determine the best network structure model. The deep learning model constructed in this paper can use real-time medical imaging of SCI patients as input for the spinal cord analysis neural network. The trained network can automatically identify and label the location of SCI, achieving a model mAP of 88.6% and an image test speed of 0.22s per image.