SDAG-CNN: Attention-Guided Convolutional Neural Network for Spinal Lesion Diagnosis and Multi-Class Classification of X-ray Radiographs

Abstract:
Spinal lesions, caused by various factors like trauma, infections, tumors, degenerative changes, and autoimmune diseases, can lead to symptoms such as pain, weakness, and loss of function. Scoliosis and spondylolisthesis are common types that cause significant discomfort and functional limitations. Early diagnosis is crucial to prevent worsening symptoms and declining quality of life. Traditional diagnostic methods, such as clinical assessments and biopsies, are limited by their reliance on individual expertise, leading to inconsistencies and delays, and may be impractical in resource-limited settings. To address these challenges, this study proposes an attentionguided convolutional neural network (CNN) for the multiclass classification of spinal lesions using X-ray imaging. The attention mechanism in the model enhances feature extraction by capturing intricate patterns and details within X-ray images, leading to more accurate and reliable identification of spinal lesions and achieving a diagnostic accuracy of 93.33%. By automating the diagnostic process and using Efficient Channel Attention blocks, the model reduces overfitting, maintains a compact size for deployment in low-computational-cost systems, and adapts to various medical imaging tasks, ultimately improving the effectiveness and efficiency of spinal lesion diagnosis. The model’s high accuracy, combined with its efficiency and accessibility, highlights its potential for significant impact in the field of medical diagnostics, particularly in enhancing the early detection and treatment of spinal lesions.