DEEP LEARNING–DRIVEN AUTOMATED SEGMENTATION AND QUANTITATIVE MRI ANALYSIS OF SPINAL STRUCTURES

Abstract

Magnetic Resonance Imaging (MRI) plays a pivotal role in the evaluation of degenerative spinal disorders due to its superior soft tissue contrast, non-invasive nature, and absence of ionizing radiation. The present study proposes an artificial intelligence–based framework for automated segmentation and quantitative measurement of key spinal structures, including intervertebral discs, vertebrae, and the spinal canal, across the cervical, thoracic (dorsal), and lumbar regions. A large-scale dataset comprising over one million MRI scans from diverse age groups, genders, and scanner manufacturers was utilized to ensure robustness and generalizability. The segmentation task was performed using the nnU-Net architecture, which automatically adapts its configuration to dataset-specific characteristics, while quantitative measurements of disc height and spinal canal anteroposterior diameter were obtained using a dedicated 3D convolutional neural network. Model performance was evaluated using the Dice coefficient and mean squared error metrics. The proposed system demonstrated high segmentation accuracy and reliable measurement precision, highlighting its potential to reduce manual workload, improve diagnostic consistency, and support clinical decision-making in the assessment and monitoring of degenerative spinal conditions.