34 Novel Intensity based Brain Abnormality Detection in MRI Head Volumes T. Kalaiselvi 1 and P. Kumarashankar 2 Assistant Professor 1 , Research Scholar 2 Department of Computer Science and Applications, The Gandhigram Rural Institute – Deemed University, Gandhigram, Tamilnadu, India. kalaiselvi.gri@gmail.com, pkumarashankar@gmail.com Abstract In the proposed work, we have created an automatic method to detect the abnormalities in brain due to tumors by analyzing the MRI head scan images. We have used the fluid-attenuated inversion recovery (FLAIR) and T2 weighted MRI images. The method uses the intensity of the images for analysis.We have used the BRATS image sets for analysis.Tumors appear hyper intense in T2 and FLAIR scan images. The cerebro spinal fluid (CSF) also appears hyper intense in the T2 scan images. We have calculated the accuracy and similarity for the input volumes. Keywords: Image mining, abnormality detection, MRI, FLAIR, T2, whole tumor mask, tumor, image processing, medical imaging 1. Introduction Scan imaging of internal body organs of human body plays a vital role in the abnormality detection of these organs. In case of human brain, imaging techniques are very helpful in identifying the abnormalities like tumors, dementia etc. Magnetic resonance imaging (MRI) is a very important technique that helps in analyzing the characteristics of the internal organs, evaluate their functions and continuously monitor the organ functions that could help in understanding the progress of treatment given. The MRI scans use radio waves and strong magnetic fields to form images of the organs. MRI is based on the nuclear magnetic resonance technique. A MRI scanner produces three types of images, namely proton density (PD) image, T1-weighted images and T2-weighted images. In addition to this we use the FLAIR technique to suppress the signals from fluids like CSF in brain. Abnormalities like tumors are shown hyper intense in T2 and FLAIR images. The tumors are shown as hypo intense or iso intense in T1 weighted scans [1, 2]. The tumors and cerebro spinal fluid appear hyper-intense in T2-weighted scans. T2–weighted images are usually very sensitive in evaluating the brain pathology; patients who are suspected to be with any type of intracranial disease are first screened with T2-weighted spin-echo and FLAIR images [3]. If there is an abnormality found in these scans then additional scans are taken in order to characterize the lesion. According to the brain screening protocol suggested by Hasselink [3], T1 images are to be taken only if T2 images show abnormalities. Tumours appear hyper-intense in T2 scans and have a similar intensity equal to that of CSF. But in FLAIR images the CSF is suppressed and hence CSF is shown as hypo intense. This plays a very important role in distinguishing the abnormalities from other brain parts. In normal slices, the CSF is symmetric about the vertical central line. Hence, if there are any abnormal tissuespresent in the CSF, the symmetry will be disturbed. This can be easily identified by measuring the symmetry of the CSF about the vertical central line [4]. Primary brain tumours could be classified into two types, namely high grade glioma (HG) and low grade glioma (LG) depending on the nature of the tumours. There is a number of automatic, semi-automatic and atlas based methods that are currently available for brain abnormality detection and segmentation. Fully automatic methods run without any manual intervention. Semi- automatic methods usually require frequent manual intervention. Atlas based methods require predefined atlas that helps to map the abnormalities in the images. Some of these methods also require multi-channel images which are not usually available in clinical routine [5-12]. There are quite a lot of methods used for automatic brain extraction from MRI scan images [13]. We propose a method that could be used on top of this existing method. Statistical value based methods are available to detect the abnormalities in the brain [14]. Such methods also use the intensity values to calculate certain statistics like Abnormality Mean and compare these against the original volume sets to identify the slices containing abnormality. In the proposed work, we are using the intensity of the MRI scan images to generate a mask of tumour in the brain MRI slices and map the slices against these tumours to identify the similarity and calculate the accuracy of the brain image slices. Our method is used on top of the existing method that extracts the brain image portion from the whole brain MRI image sets. We have used MRI FLAIR and T2 images of varying sizes. The image sets are of varying dimensions. Computational Methods, Communication Techniques and Informatics ISBN: 978-81-933316-1-3 34