Spinal Cord Segmentation and Injury Detection based on Siamese Conventional WideRes Network using CT Image.

Spinal Cord Injury (SCI) presents a complex clinical challenge, frequently leading to a cascade of secondary health issues. These may include impaired temperature regulation, sudden drops in blood pressure upon standing, episodes of Autonomic Dysreflexia, and disruptions in urinary and gastrointestinal function. With the surge of machine learning, multiple schemes have been proposed for disease diagnosis using medical imaging, even for SCI detection. However, issues like class imbalance and variability in imaging affect the performance of these models. Hence, this research presents a novel Deep Learning (DL) method for spinal cord segmentation and injury detection, employing a Siamese Convolutional WideRes Network (SCWRes-Net) with Computed Tomography (CT) images. The input CT images are initially accumulated from the dataset and processed for spinal cord segmentation employing a Mask Regional Convolutional Neural Network (MRCNN). After the segmentation, the active contour approach is utilized for disc localization. Then, the extraction of features is performed to obtain features, like geometric characteristics, connectivity, and image-level features. Finally, SCI detection is executed employing the SCWRes-Net model, which integrates the Siamese Convolutional Neural Network (SCNN) and Wide Residual Network (WideResNet). The SCWRes-Net model demonstrated impressive performance, achieving an accuracy of 92.56%, a True Negative Rate (TNR) of 91.80%, and a True Positive Rate (TPR) of 93.33%.