A framework for segmentation and estimation of intracranial measurements in CT scans

Abstract

Inter-observer variability in assessment of medical images is considered 'Achilles heel' amongst radiologists because it can lead to missed diagnoses and grave consequences. Measurements on neuro-images to ascertain severity and extent of the pathology or trauma are routinely performed, however, poor perception, inaccurate deduction, incomplete knowledge or the quality of the image can affect the intuition of the doctor leading to errors and variation. In this paper we present a novel, hybrid technique for segmentation of significant anatomical landmarks using template matching, artificial neural networks and level sets, and estimation of various ratios and indices as well as haematoma volume on brain CT scans. The proposed method is efficient and robust in segmenting cross-sectional, noncontrast CT scans and has been evaluated on images from subjects with different ages and both genders. The proposed method has an average ICC ≥ 0.97 and Jaccard Index ≥ 0.86 compared with the experts. Hence, our approach can be used in processing data for further use in research and clinical environment to provide second opinions very close to the experts' intuition.