Automatic detection of brain tumors using MRI images

Objective: This research has been performed to determine the regions of tumors in brain MR images using segmentation methods. Material and Methods: The methods as K-Means, Fuzzy C-Means, Self Organizing Maps and Otsu have been used to detect the border regions of brain tumors in the brain MR image dataset. In addition, the hybrid K-Means+Otsu method has been developed from the combination of K-Means and Otsu methods which are the most successful in our systems. With these methods, the segmentation applications have been applied on the brain tumor images. Before the segmentation, the skull parts of the images have been removed from the brain image with the developed skull removing method. Furthermore, some restorations on the brain image have been made with median, Gaussian and Wiener filters. Results: To segment the brain image, some systems using K-Means, Fuzzy C-Means, Self Organizing Map, Otsu and hybrid K-Means+Otsu methods were developed. The success of these systems was compared with accuracy, sensitivity and specificity measures. With and without skull removal process, the success of the systems has been evaluated. After the skull removal, approximately 10% better results were obtained. The hybrid method was the most successful with 94% accuracy measure while the K-Means was the second with 93%. According to the sensitivity measure, the hybrid method has been the most successful with 94%. However, both K-Means and the hybrid method were the most successful with 93% by the measure of specificity. Conclusions: This study showed that the success rates have highly increased after the skull removal in brain MRI images. According to the results obtained with the segmentation methods, the hybrid K-Means+Otsu method has been more successful than the other methods.