Multi-class Multi-label Classification and Detection of Lumbar Intervertebral Disc Degeneration MR Images using Decision Tree Classifiers

Abstract

Evidence-based medicine decision-making based on computer-aided methods is a new direction in modernhealthcare. Data Mining Techniques in Computer-Aided Diagnosis (CAD) are powerful and widely used toolsfor efficient and automated classification, retrieval, and pattern recognition of medical images. They becomehighly desirable for the healthcare providers because of the massive and increasing volume of intervertebral discdegeneration images. A fast and efficient classification and retrieval system using query images with high degreeof accuracy is vital. The method proposed in this paper for automatic detection and classification of lumbarintervertebral disc degeneration MRI-T2 images makes use of texture-based pattern recognition in data mining.A dataset of 181segmented ROIs, corresponding to 89 normal and 92 degenerated (narrowed) discs at differentvertebral level, was analyzed and textural features (contrast, entropy, and energy) were extracted from each disc-ROI. The extracted features were employed in the design of a pattern recognition system using C4.5 decisiontree classifier. The system achieved a classification accuracy of 93.33% in designing a Multi-class Multi-labelclassification system based on the affected disc position. This work combined with its higher accuracy isconsidered a valuable knowledge for orthopedists in their diagnosis of lumbar intervertebral disc degeneration inT2-weighted Magnetic Resonance sagittal Images and for automated annotation, archiving, and retrieval of theseimages for later on usage.Keywords: Data Mining, Image Processing, Lumbar Intervertebral Disc Degeneration, MRI-T2, Decision Trees,Multi-class Multi-label Classification.