This work focuses on fully automatic detection of brain tumors. The first aim is to determine, whether the image contains abrain with a tumor, and if it does, localize it. The goal of this work is not the exact segmentation of tumors, but the localization oftheir approximate position. The test database contains 203 T2-weighted images of which 131 are images of healthy brain and theremaining 72 images contain brain with pathological area. The estimation, whether the image shows an afflicted brain and where apathological area is, is done by multi resolution symmetry analysis. The first goal was tested by five-fold cross-validation techniquewith 100 repetitions to avoid the result dependency on sample order. This part of the proposed method reaches the true positive rateof 87.52% and the true negative rate of 93.14% for an afflicted brain detection. The evaluation of the second part of the algorithmwas carried out by comparing the estimated location to the true tumor location. The detection of the tumor location reaches therate of 95.83% of correct anomaly detection and the rate 87.5% of correct tumor location.


Dvořák, P., Kropatsch, W. G., & Bartušek, K. (2013). Automatic Brain Tumor Detection in T2-weighted Magnetic Resonance Images. Measurement Science Review, 13(5), 223–230. https://doi.org/10.2478/msr-2013-0034