Hybrid deep learning model for spinal tumor diagnosis on MRI scans.
Authors
Affiliations (4)
Affiliations (4)
- Department of Biomedical Engineering, Excel Engineering College, Namakkal, India.
- Department of Electronics and Communication Engineering, Gnanamani College of Technology, Namakkal, India.
- Department of Electronics and Communication Engineering, Kalaignar Karunanidhi Institute of Technology, Coimbatore, India.
- Department of Electronics and Communication Engineering, Excel Engineering College, Namakkal, India.
Abstract
BackgroundDiagnosing spinal tumors (ST) has always been challenging, especially when distinguishing between benign and malignant types. Incorrect diagnosis can lead to inappropriate treatment plans for patients. Previous studies have primarily focused on detecting and classifying these tumors using MRI scans. However, further research is needed to improve diagnostic accuracy.ObjectiveThis study proposes a novel approach that incorporates both MRI images and patient age data to enhance the detection and classification of spinal tumors. The approach utilizes Inception V3 for local feature extraction and the Vision Transformer (ViT) for global feature classification, addressing long-term dependencies in tumor data.MethodMRI images are pre-processed using an Average Filter (AF) and a Morphological Operator (MO) to smooth the images and convert them into binary format. The tumor detection is performed using a hybrid deep learning model, integrating age-related information to improve classification accuracy.ResultsThe pre-processed data is passed through a Self-Attention Fusion Mechanism (SAFM) to refine the results and enhance diagnostic accuracy by filtering out irrelevant information. The model's performance is evaluated through various metrics, including accuracy, sensitivity, and specificity, showing significant improvements over existing techniques.ConclusionThe proposed model demonstrates superior accuracy in diagnosing spinal tumors by combining MRI imaging and patient age data. It effectively differentiates benign from malignant tumors, providing a reliable tool for clinicians. The model achieved a specificity of 96%, accuracy of 93%, and computational delay of just 10.04 s, outperforming existing diagnostic models.