Discrimination Of Various Brain Pathological Lesions Non-invasively by MRI Utilizing Supervised Machine Learning Manipulating Tissue Refractive Index, T2 Relaxation Values and Tissue Metabolites

Abstract

Objective: Biopsy is mandatory for final diagnosis of brain pathologies. For this, tissue is harvested by Stereotaxic or endoscopic biopsy which are hazardous and sometime life threatening. For this purpose a noninvasive method has been innovated.

Introduction: Attempt has been made to discriminate benign and malignant lesions of brain noninvasively from the T2 weighted MR image. Various data like Refractive Index(RI), T2 relaxation value, Apparent diffusion coefficient (ADC) values, metabolites of brain tissue determined by MR spectroscopy were assigned with the T2 weighted image (Ground truth image). These data after proper training by Support vector Machine (SVM) and or by Artificial Neural network (ANN), live predicting of the tissue could be possible. Principal Component Analysis (PCA) can also clusters the diseases.

Materials and Methods: RI of biopsy tissues were determined by Abbe refractometer. A relationship of RI of tissues and T2 values was established in MRI. T2 and RI map generated from the T2 weighted image with the help of RI and T2 shade. A false color RI shade helps in converting T2 image into color coded RI image characterizing benign and malignant tissue. ANN and SVM after proper training of the data can predict the diseases. PCA reduces the dimension of the data.

Results: From the relationship of RI and T2 values color coded map of the T2 weighted image discriminates various benign from malignant tissue. ANN and SVM live predict the disease. PCA diminishes the dimension of the data.

Conclusion: Almost 95% of diagnostic accuracy can be obtained non-invasively by RI color coded map and supervised machine learning.