Craniopharyngioma Detection and Segmentation in MRI Images

Abstract

A tumour is an abnormal growth of human body tissues. Tumours are classified as benign or malignant. Malignant tumours cause serious health complications that may threaten a patient's life. The diagnosis of such tumours requires experienced and trained medical specialists. Alternatively, computerised tumour detection and localisation can help physicians to reach accurate, fast and reliable diagnosis. Craniopharyngioma (CP) is a brain tumour located in the sellar and parasellar regions of the central nervous system. It causes various symptoms such as headaches, visual and neurological disturbances, growth retardation and delayed puberty. In addition to histological examinations, multiple tissue characteristics are evaluated for accurate diagnosis of CP tumours. Patients with craniopharyngiomas are treated by total excision and post-operative radiotherapy in cases that have no hypothalamic invasion or sub-total resection. Early detection and diagnosis of the tumour can minimise the complications associated with surgical and radiotherapy treatments. In this article, an image processing technique for the segmentation and detection of brain tumours in general and craniopharyngioma in particular using MRI brain images, is presented. The technique is based on K-means clustering, multiple thresholding and iterative morphological operations. It was tested on 104 MRI images and the quantitative analysis of its effectiveness showed performance values of 98%, 93%, 100%, 95% and 100% for precision, recall, specificity, Dice score eoefficient and accuracy, respectively.