ShuffleLeNet architecture for spinal cord injury classification and level detection.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Radiology

Background:

• Spinal cord injuries (SCI) require early diagnosis for effective treatment.
• Automated analysis of CT images for SCI diagnosis presents challenges in noise, segmentation, and feature extraction.
• Current methods may lack computational efficiency and diagnostic accuracy.

Purpose of the Study:

• To develop an automated and accurate framework for classifying spinal cord injuries (SCI) and detecting injury levels from CT images.
• To enhance early diagnosis and timely clinical intervention for SCI.
• To address challenges in noise sensitivity, segmentation accuracy, feature extraction, and computational efficiency in SCI imaging.

Main Methods:

• Image preprocessing using Wiener filtering for noise reduction and quality enhancement.
• Spinal cord segmentation using a modified U-Net architecture.
• Extraction of discriminative features including ImLGIP, deep features (ResNet, VGG16), and PHOG.
• Classification and injury level detection using a hybrid Improved ShuffleLeNet model.

Main Results:

• The Improved ShuffleLeNet model achieved superior performance over conventional methods.
• Achieved a classification accuracy of 0.962, a negative predictive value (NPV) of 0.972, and a precision of 0.935.
• Demonstrated effectiveness in accurately classifying spinal cord injuries and identifying injury levels.

Conclusions:

• The developed framework offers a robust and efficient solution for automated SCI diagnosis from CT images.
• Integration of preprocessing, segmentation, feature extraction, and hybrid deep learning enhances diagnostic accuracy and computational efficiency.
• The Improved ShuffleLeNet model shows potential for real-time clinical applications in SCI classification and injury level detection.