An Automated Method for Segmenting Brain Tumors on MRI Images

Abstract

We present a method for segmenting images of tumors on MRI images of the brain based on an algorithm developed for automated determination of segmentation and outlining thresholds. Testing was performed by generating two databases of real MRI images of the brain, with radiology reports. Criteria for assessment of the quality of the segmentation results were: the Dice score, the Jaccard index, sensitivity, and specificity. Analysis of results obtained using this algorithm to solve the brain tumor MRI image segmentation task showed levels of sensitivity and specificity of 89% to 99%, which is evidence that assessment of the position and boundaries of brain pathology is highly effective.