Detection and localization of brain tumors using Fractional Hartley Transform and adaptive neuro-fuzzy inference system classification methods

Brain tumors present the most unique challenge to clinicians in terms of their detection and assessment. The identification of brain lesions and tumors are often performed by magnetic resonance imaging (MRI) diagnostics. This study proposes a machine learning technique to detect and localize brain tumors via MRI of the brain with a two-phased approach. Phase one encompassed the preprocessing of the MRI images to improve the image contrast and then apply the Fractional Hartley Transform method to the preprocessed MRI. The coefficients of this transformation were used to construct the feature vectors that were used to classify a given MRI of the brain as normal or abnormal based on the adaptive neuro-fuzzy inference system (ANFIS) approach. Phase two corresponded with the localization of the regions with tumors in the abnormal brain images with the aid of morphological operations. The entire process was evaluated using multiple performance metrics, i.e., sensitivity, specificity, accuracy, and classification rate. Without considering the clinical dataset, the study results achieved corresponded to 99.4% of sensitivity, 99.2% of specificity, and 99.3% of accuracy; however, when considering the brain tumor image segmentation (BRATS) 2015 clinical dataset, the results achieved corresponded to 97.24% of sensitivity, 97.98% of specificity, and 97.54% of accuracy in the tumor detection process. These approaches are of extreme importance in the automated estimation of tumors via MR brain images and aid the clinicians to determine a precise clinical diagnosis, using robust machine learning techniques to mitigate human errors.