Open AccessEfficient Clustering of Unlabeled Brain DICOM Images based on similarity
Author(s) -
Suriya Murugan,
M. G. Sumithra,
M. Murugappan
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1916/1/012017
Subject(s) - cluster analysis , dicom , computer science , artificial intelligence , silhouette , pattern recognition (psychology) , similarity (geometry) , feature extraction , computation , computer vision , feature (linguistics) , medical imaging , image (mathematics) , linguistics , philosophy , algorithm
Clustering has proven to be an effective method in the medical field for finding patterns in labelled and unlabelled datasets. This work is implemented over whole body CT scans (∼1TB) of 3500 patients in form of unlabelled DICOM images. The whole-body CT images have been anonymized for 30 attributes based on DICOM regulations and the Brain images alone are segmented using the DICOM tag element called ‘Protocol stack’. The segmented Brain images are efficiently grouped based on visual similarity using K-means clustering after performing feature extraction and dimensionality reduction. The results of the clustering can be furtherutilized by radiologists to perform labelling or find patterns in Brain CT scans of patients that are difficult where each scan consists of a varying number of slices during detection of Internal Bleeding. The efficiency of K-means is analyzed by performing computation over a different number of clusters (K) by applying silhouette scores to find optimal cluster.