Back to all papers

EADC-Net: A Dual-Stage Deep Learning Framework with Spatial and Channel Central Attention for Automated Detection of Lumbar Degenerative Disc Disease from Radiographs.

September 28, 2026pubmed logopapers

Authors

Chang CC,Chen YY,Fang TL,Chen KW,Yang YC

Affiliations (8)

  • Spine Center, Taichung Municipal Geriatric Rehabilitation General Hospital-Managed By China Medical University, Taichung, Taiwan.
  • Department of Orthopaedics, Taichung Municipal Geriatric Rehabilitation General Hospital-Managed By China Medical University, Taichung, Taiwan.
  • Department of Leisure Industry Management, National Chin-Yi University of Technology, Taichung, Taiwan.
  • School of Medicine, China Medical University, Taichung, Taiwan.
  • Artificial Intelligence and Computer Engineering Department, National Chin-Yi University of Technology, Taichung, Taiwan. [email protected].
  • Information Management Department, National Chung Hsing University, Taichung, Taiwan.
  • Faculty of Engineering and Information Technology, The University of Melbourne, Melbourne, Australia.
  • Faculty of Science, The University of Melbourne, Melbourne, Australia.

Abstract

Lumbar degenerative disc disease (DDD) is a primary driver of chronic back pain and spinal deformity; however, MRI-based diagnosis is constrained by high costs and limited accessibility. We introduce EADC-Net, an adaptive deep learning framework for automated DDD detection from standard radiographs to facilitate early clinical screening. EADC-Net utilizes a dual-stage architecture. Lumbar regions of interest are initially extracted via segmentation deep residual neural network (SRNet), followed by contrast-limited adaptive histogram equalization (CLAHE) contrast enhancement. To account for disc height interdependencies, images are partitioned into overlapping segments (L1-L3, L2-L4, and L3-L5). The detection engine, built on a VGG-based backbone, incorporates a novel spatial and channel central attention (SCCA) module. This module synergistically leverages global weighted average pooling (GWAP) and coordinate attention (CA) to prioritize features in the nucleus pulposus and disc boundaries. Evaluated on 402 radiographs from 204 patients under a rigorous patient-level fivefold cross-validation framework, EADC-Net achieved an average accuracy of 86.26%, recall of 79.14%, precision of 81.00%, and an F1-score of 79.79%. Comparative analysis and ablation studies confirmed that EADC-Net significantly outperforms benchmark architectures by effectively capturing subtle degenerative features. EADC-Net offers a reliable, cost-effective tool for automated DDD detection. By bridging the gap between radiographic imaging and high-accuracy deep learning, this framework supports early spinal health management.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.