Deep Learning for Diagnosis of Disc Herniation in Small Animals: A CNN-Based Approach Using CT Imaging

Intervertebral disc disease (IVDD) is a common neurological disorder in dogs and cats, leading to significant mobility issues and reduced quality of life. Traditional imaging modalities, including radiography, computed tomography (CT), and magnetic resonance imaging (MRI), are integral to IVDD diagnosis, but each has limitations in visualizing soft tissues and requires specialized interpretation. Recent advancements in deep learning (DL), particularly convolutional neural networks (CNNs), offer promising solutions by automating image analysis, enhancing diagnostic accuracy, and reducing human error. This study presents the development of a CNN-based model for detecting disc herniation in CT images of small animals. A total of 1651 annotated CT images were curated from 64,602 scans to create a balanced dataset of healthy and herniated disc cases. Data preprocessing included image resizing, normalization, and extensive augmentation to improve generalization. The segmentation of intervertebral discs was performed using a U-Net architecture, supplemented by synthetic mask generation with K-means clustering and contrast enhancement via Contrast Limited Adaptive Histogram Equalization (CLAHE). Classification employed a fine-tuned Visual Geometry Group 16 (VGG16) model with transfer learning. This approach incorporated optimized layers for binary classification, leveraging dropout and batch normalization to prevent overfitting. The model demonstrated high classification accuracy, providing rapid and objective analysis of CT scans for IVDD diagnosis. By integrating DL with veterinary imaging, our research advances diagnostic capabilities, aiming for earlier detection and better therapeutic planning for small animals with spinal disorders. Future work includes expanding datasets and refining the model for broader clinical applicability.