Machine learning classificatory as a tool in the diagnosis of amyotrophic lateral sclerosis using diffusion tensor imaging parameters collected with 1.5T MRI scanner: A case study.

The relevance of the study lies in the need to improve the diagnosis of amyotrophic lateral sclerosis (ALS) by utilizing diffusion tensor imaging (DTI) obtained from conventional 1.5 Tesla MRI scanners. The study aimed to investigate the potential of using different machine learning (ML) classifiers to distinguish between individuals with ALS. In this study, five ML classifiers ("support vector machine (SVM)", "k-nearest neighbors (K-NN)", naïve Bayesian classifier, "decision tree", and "decision forest") were used, based on two DTI parameters: fractional anisotropy and apparent diffusion coefficient, obtained from two manually selected ROIs at the level of the brain pyramids in 47 ALS patients and 55 healthy subjects. The quality of each classifier was evaluated using the confusion matrix and ROC curves. The highest accuracy in differentiating ALS patients from healthy individuals based on DTI data was demonstrated by the radial kernel support vector method (77% accuracy [p=0.01]), while K-NN and "decision tree" classifiers had slightly lower performance, and "decision forest" classifier was overtrained to the training set (AUC=1). The authors have shown a sufficiently accuracy of ML classifier "SVM" in detecting radiological characteristics of ALS in pyramidal tracts. [ABSTRACT FROM AUTHOR]