Automatic Diagnostic System of Lumbar Disc Herniation MRI Images Based on CBAM-YOLO

LI Yahao; SHEN Xueqiang; JIANG Hong; YU Pengfei

Abstract：Objective Lumbar disc herniation(LDH)is a common type of degenerative spinal disease that can lead to low back pain and neurological symptoms in the lower limbs.MRI images are crucial in diagnosis but suffer from empirical dependence and lack of standardisation mechanisms.The aim of this study was to develop a deep learning model based on CBAM-YOLO to assist in the automated diagnosis of lumbar disc herniation MRI images in order to improve the accuracy and efficiency of diagnosis.Methods A publicly available dataset containing MRI image data of 643 LDH patients was adopted in this study,and each image was labelled in detail with a variety of structural information including intervertebral discs and others.This study proposes a novel model,CBAM-YOLO,which is improved from the original YOLO-v8 model by embedding the convolutional attention module CBAM.This improvement helps the model to identify the feature location and spatial distribution information of the herniated disc more accurately.In order to fully validate the performance of the model,this study used the CBAM-YOLO model to systematically train the data-enhanced training set for a total of 100 training cycles.In the model evaluation session,this study used several evaluation metrics such as Precision,Recall,F1-Score,Accuracy,and mAP in order to comprehensively and rigorously assess the performance of the model.Results The CBAM-YOLO model demonstrated excellent performance in diagnosing lumbar disc herniation MRI images.Compared with the original YOLO-v8 model,the CBAM-YOLO model constructed in this study showed significant advantages in Precision of 89.9%,Recall of up to 100.0%,F1-Score of 94.6%,Accuracy of 89.9%,and mAP of 97.1%.Conclusion The test results fully highlight the great potential of deep learning-based automated diagnostic systems for disc identification and segmentation in lumbar spine MRI images,as well as the potential to improve the accuracy and efficiency of disease diagnosis in clinical applications,thereby reducing the burden on healthcare professionals.