Related Experiment Videos
Automated Classification of Cervical Spinal Stenosis Using Deep Learning on Computed Tomography Scans
Yu-Long Zhang1, Jia-Wei Huang, Kai-Yu Li
1Department of Spine Surgery, Zhejiang Spine Research Center, The Second Affiliated Hospital and Yuying Children's Hospital of Wenzhou Medical University, Wenzhou, Zhejiang, China.
Spine
|June 3, 2025
Summary
A new deep learning (DL) model using computed tomography (CT) images accurately diagnoses cervical spinal stenosis (CSS). This AI tool shows diagnostic performance comparable to experienced physicians, offering a valuable alternative to MRI.
Area of Science:
- Radiology
- Artificial Intelligence
- Medical Imaging
Background:
- Magnetic resonance imaging (MRI) is standard for cervical spinal stenosis (CSS) diagnosis but has limitations.
- Computed tomography (CT) is a viable alternative, especially when MRI is contraindicated or unavailable.
- CT-based deep learning (DL) models can enhance diagnostic accuracy beyond conventional CT.
Purpose of the Study:
- To develop and validate a DL model for diagnosing CSS using CT images.
- To assess the model's diagnostic performance against human physicians.
Main Methods:
- A retrospective study using paired CT/MRI images.
- A two-stage DL model: Faster R-CNN for region identification and CNNs for classification.
- Model evaluation using accuracy, F1-score, and Cohen κ coefficient.
Main Results:
- The DL model achieved high accuracy in both multiclass (up to 89.40%) and binary (up to 96.03%) CSS classification.
- Model consistency with senior physicians exceeded 90% in binary classification tasks.
- Performance demonstrated significant diagnostic capability, comparable to experienced radiologists.
Conclusions:
- The developed DL model accurately analyzes CT images for CSS diagnosis.
- The model's performance rivals that of senior physicians, indicating its clinical utility.
- CT-based DL offers a promising, efficient tool for CSS diagnosis.
