Related Experiment Video
Updated: Aug 19, 2026
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
IShuffleNet-PCNN hybrid classifier within an MResU-Net-based framework for spinal cord injury detection and
N Chiranjeevi1, S Shafiulla Basha2
1Y.S.R Engineering College of Yogi Vemana University, Proddatur, Andhra Pradesh, India. chiranjeevin586@gmail.com.
Background:
Spinal cord injury (SCI) is a serious medical condition. Spinal Cord Injury limits the movement of the body, blocks the nervous system and affects the quality of life of an injured patient. Accurate detection and classification of these fractures are essential for timely diagnosis and treatment planning; however, conventional assessment methods often struggle with noise, variability, and subtle injury patterns in CT imaging.This study aimed to develop an integrated deep learning framework for accurate and robust classification of spinal cord injury-related fractures from CT images.
Methods:
The proposed framework consisted of preprocessing, segmentation, feature extraction, and classification stages. In the preprocessing stage, a Weighted Balanced Anisotropic Filtering (WBAF) method was used to reduce CT image noise while preserving structural information. Spinal cord regions were subsequently segmented using a Modified Residual U-Net (MResU-Net). An Improved Pyramid Histogram of Oriented Gradients (IPHOG) method incorporating Gaussian smoothing and adaptive weighting was then employed to extract discriminative structural and fracture-related features. Finally, the extracted features were classified using the proposed IShuffleNet-Parallel Convolutional Neural Network (IShuffleNet-PCNN), incorporating Group Normalization and Adaptive Swish-Mish activation to improve training stability and feature discrimination.
Results:
The proposed IShuffleNet-PCNN achieved an accuracy of 0.933, precision of 0.937, and negative predictive value (NPV) of 0.991, demonstrating effective classification performance. The combination of WBAF, MResU-Net, IPHOG, and IShuffleNet-PCNN provided a comprehensive pipeline for noise reduction, anatomical segmentation, discriminative feature extraction, and spinal cord injury classification.
Conclusion:
The proposed deep learning framework demonstrated promising performance for CT-based spinal cord injury classification. The integration of noise-preserving preprocessing, modified segmentation, enhanced feature extraction, and the IShuffleNet-PCNN classifier may improve automated fracture assessment and provide potential support for timely clinical decision-making.