BINARY CLASSIFICATION OF SPINAL CORD INJURY USING ENSEMBLE MACHINE LEARNING VOTE ALGORITHMS

Abstract

Spinal cord injury (SCI) is a clinically pertinent disorder which is needed to be evaluated in a timely manner. The proposed research is based on the machine learning model used for binary classification of SCI using MRI images. There are four segmentation (Fuzzy C-Means, Region Growing, K-Means and Expectation Maximization) that were evaluated and K-Means was the best performer due to its effectiveness in identifying the spinal cord structures. GLCM texture attributes were also extracted on the segmented areas where 14 features had been extracted and narrowed down to 8 features to be applied in the categorization. Three classical machine learning models were built and tested on the identical set of features. The ML models utilized were: 1) K-Nearest Neighbour (KNN), 2) Probabilistic Neural Network (PNN) and 3) Naive Bayes (NB). It was found that KNN was the most balanced to the overall performance. The findings indicate that conventional classifiers, with the application of the features based on textures, could be useful in assisting the detection of automated SCI. The study provides a rational and understandable basis that future solutions grounded on hybrids or ensembles can be created with the aim to improve the accuracy of diagnosis.