Deep Learning Approaches for Lumbar Spine MRI Segmentation and Classification

Lumbar spine disorders are prevalent conditions that can lead to chronic pain and reduced quality of life, often requiring accurate diagnosis through magnetic resonance imaging.However, manual interpretation of MRI scans is time-intensive and may lead to diagnostic errors due to the complexity of spinal anatomy and subtle differences between healthy and pathological tissues.This study aims to develop an automated method for segmenting and classifying the lumbar spine, including the assessment of lumbar lordosis, using deep learning techniques.In the first stage, spinal segmentation is performed using a combination of image thresholding and a convolutional neural network based on the U-Net architecture, with binary conversion of ground truth images and data augmentation techniques to enhance the dataset.The data is divided into training and testing sets, with 80% allocated for model training.In the second stage, classification of lumbar lordosis is achieved through a method called "Deteclordose" and transfer learning, following the automatic calculation of the Cobb angle.Images are then categorized into three classes: hyper lordosis, hypo lordosis, and normal lordosis.The method was evaluated on a dataset of MRI images obtained from Jordanian hospitals, achieving promising results, including a segmentation accuracy of 99.30%, a loss value of 0.025, a classification accuracy of 96%, a recall of 94%, and an F1 score of 93%.These outcomes demonstrate the potential of the proposed approach to improve diagnostic accuracy and reduce the burden on radiologists by providing a reliable, automated system for analyzing lumbar spine MRI images.