Mutual Information-Based Optimum Metrics Identification in Symmetry-Based Brain Abnormality Detection

Abstract

Abstract In our previous studies, we showed that brain abnormalities can be detected by comparing the features extracted from the two lobes with each other. Based on this, many metrics, such as those extracted from colour or texture features, have been extracted and used. The large number of extracted metrics posed a challenge in terms of how important each metric is. In this research, we use the mutual information content to measure the importance of the metrics and their influence on the classification process as it gives an indication of how the output and each input are related to each other. The algorithm was applied to 366 images, from which eleven metrics were extracted and studied. Random forest classifier was used as it was proven that it gives the highest accuracy. The obtained results showed that 30% of the features can be eliminated without a significant effect on the accuracy.