Application of modified balanced iterative reducing and clustering using hierarchies algorithm in parceling of brain performance using fMRI data

Abstract

Introduction: Clustering of human brain is a very useful tool for diagnosis, treatment, and tracking of brain tumors. There are several methods in this category in order to do this. In this study, modified balanced iterative reducing and clustering using hierarchies (m-BIRCH) was introduced for brain activation clustering. This algorithm has an appropriate speed and good scalability in dealing with very large data using a new concept of Clustering Feature. Materials and Methods: In this study, data from the brain scan had been used. This dataset consisted of 74 consecutive brain scans. After data preprocessing, brain scan images were clustered through the BIRCH and m-BIRCH algorithms. Data were analyzed using WFU-PickAtlas in Matlab software and were compared with the TD Lobes Standard Atlas. Results: The speed of implementation of the m-BIRCH algorithm decreased as threshold limit increased. The m-BIRCH clustering algorithm showed that there was no specific ascending or descending pattern between branch factor and the run-time of the algorithm. The maximum runtime value of the algorithm was related to the branching factor of 30 which was 94 seconds, equivalent to the upper threshold limit of the BIRCH algorithm. Conclusion: Applying the m-BIRCH algorithm on high-dimension data set such as brain scan images has relative advantages and provides a tradeoff between time and space complexity. By simultaneously increasing the branching factor and threshold limit, the sensitivity of clustering will be decreased