CT brain image clustering for differentiation of intracerebral haemorrhage: a novel algorithm.

Abstract

Background and Aims: Hypertensive haemorrhage (HH) and cerebral amyloid angiopathy haemorrhage (CAAH) are difficult to differentiate on clinical grounds, although there are distinguishing imaging features. The gold-standard, MRI, can differentiate between more obvious cases but CT remains the preferred modality due to accessibility and cost, especially in developing economies. Method: In this study, CT brain images (n=40) of primary ICH were acquired from the UKM Medical Centre, Imaging Centre Patient Database. The DICOM images were anonymized to remove the metadata and resized into 256 x 256 pixels to standardize image resolution. Skull removal and normalization were required in further processing of the images to improve identification of the ICH bleed location through the k-means clustering algorithm. Results: K- means clustering identified 7 clusters, namely; grey matter (2 clusters), white matter, cerebrospinal fluid (2 clusters) and hematoma (2 clusters), as visualized in Figure 1. The significance of the extracted clusters will be analysed using texture feature extraction. Further, the Support Vector Machine (SVM) classifier algorithm will be used to classify the different type of ICH bleed using the significant features. Conclusion: The proposed segmentation technique is expected to help differentiate between HH and CAAH on CT.