A Novel Textural and Morphological-Based CAD System for Early and Accurate Diagnosis of Vertebral Tumors

Abstract

Spinal tumors can rapidly lead to disability and even death. Cancerous vertebral tumors (VTs) represent 90% of spine cancer. In this paper, a novel computer-aided diagnostic (CAD) system with the ability to provide an early, accurate, and non-invasive diagnosis of VTs is suggested. The proposed VTs-CAD integrates different combinations of textural and morphological imaging markers extracted from T1 and T2-MRI. A total of 47 patients with biopsy proven VTs (23 benign & 24 malignant cases) underwent both types of MRI scans and provided their consent to participate in this study. VTs-CAD applied an adaptive distance-maps algorithm on the segmented tumor to obtain a new region of interest (ROI) of the tumor and its surrounding tissues to capture the infiltrative effect of cancerous tumors on surrounding healthy tissues. 46 textural markers were extracted from ROIs. A spherical harmonic model was used to extract morphological markers from the tumor itself. The VTs-CAD was evaluated using different combinations of these markers along with various machine learning algorithms. Optimal combination of the extracted markers was reported by the SVM classifier with a Gaussian kernel. It achieved 93.6% accuracy, 91.7% sensitivity, 95.7% specificity, and 93.6% F1-score. These findings suggest that the VTs-CAD can be effectively utilized to diagnose VTs using T1/T2-MRIs accurately and non-invasively at an early stage. © 2023 IEEE.

Author keywords Machine Learning; Morphological Markers; Multimodal MRIs; Textural Markers; VTs-CAD