Neuro-levelset system based segmentation in dynamic susceptibility contrast enhanced and diffusion weighted magnetic resonance images Vijayakumar Chinnadurai a,n , Gharpure Damayanti Chandrashekhar b a Department of Radio diagnosis and Imaging, Armed Forces Medical College, Pune, Maharashtra 411040, India b Department of Electronic Science, University of Pune, Pune, Maharashtra, India article info Article history: Received 17 June 2011 Received in revised form 24 November 2011 Accepted 29 February 2012 Available online 9 March 2012 Keywords: Neuro-levelset method Artificial neural networks Radial basis function Self-organizing map Dynamic contrast susceptibility magnetic resonance images Diffusion weighted images abstract In this study, neuro-levelset method is proposed and evaluated for segmentation and grading of brain tumors on reconstructed images of dynamic susceptibility contrast (DSC) and diffusion weighted (DW) magnetic resonance images. The proposed neuro-levelset method comprises of two independent phases of processing. At first, reconstructed images have been independently processed by three different artificial neural network systems such as multilayer perceptron (MLP), self-organizing map (SOM), and radial basis function (RBF). The images used for these tasks were the cerebral blood volume (CBV), time to peak (TTP), percentage of base at peak (PBP) and apparent diffusion coefficient (ADC) images. This processing step ensued in formation of segmentation images of brain tumors. Further, in the second phase, these coarse segmented images of each artificial neural network system have been independently subjected as speed images to levelset method in order to optimize the segmentation performance. This has resulted in construction of three distinct neuro-levelset methods such as MLP-levelset, SOM-levelset and RBF-levelset method. Proposed neuro-levelset methods performed better in segmenting tumor, edema, necrosis, CSF and normal tissues as compared to independent artificial neural network systems. Among three neuro-levelset methods, RBF-levelset system has performed well with average sensitivity and specificity values of 91.43 72.94% and 94.43 71.90%, respectively. Crown Copyright & 2012 Published by Elsevier Ltd. All rights reserved. 1. Introduction Magnetic resonance imaging engages a prominent task in diagnosis and grading of brain tumors, and it is now one of the essential modalities for adequate clinical management of many tumor types. However, there are yet many challenges and difficulties in differentiating brain tumors and their pathologies. Inability of distinguishing edema and necrosis from the tumor is one of its primary limitations. In order to circumvent these inadequacies, many advanced techniques have been developed over the years. Some of the most important developments are evolution of newer image segmentation techniques and image acquisition techniques. Dynamic susceptibility contrast (DSC) studies and diffusion-weighted MRI (DWI) studies are recent advances in image acquisition techniques, which depict tumor morphology and relationships of malignant lesions to neighboring structures better. MR diffusion study [1] is sensitive to the molecular diffusion of water, has been well substantiated as a reliable non-invasive approach in distinguishing necrotic spaces associated with malignant brain tumor from the benign ones. Contemporary advancements in dynamic susceptibility contrast (DSC) techniques [2, 3] have allowed the formation of cerebral blood volume (CBV), time to peak (TTP), percentage of base at peak (PBP) images, which lead to the qualitative and quantitative inspection of tumor vascularity. These images have eased in the assessment of tumor grade and in choosing the right site of biopsy. DWI and DSC in non-enhancing brain tumors offer clinically relevant physiological data, which is otherwise not realizable by conventional T1 and T2 MRI. Non-enhancing brain tumor with lower ADC values in the solid regions and higher CBV ratios in both solid portion and peritumoral regions of tumors are significantly correlated with anaplasia. Therefore, DWI and DSC have been unified for providing better feature information in the auto- matic image segmentation task of brain tumor in order to predict tumor grading in this study. However, diffusion anisotropy based techniques such as diffusion tensor imaging and fractional aniso- tropy have not been included due to its least influence in predicting the grades of the brain tumor. Brain tumor segmentation [4] and grading in DSC and diffusion weighted (DW) MR Images are, nevertheless, a challenging issue due to the complicated appearance of tumors in these images and their irregular sizes and boundary. Some brain tumors also distort other structures and appear together with edema that hurdles Contents lists available at SciVerse ScienceDirect journal homepage: www.elsevier.com/locate/pr Pattern Recognition 0031-3203/$ - see front matter Crown Copyright & 2012 Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.patcog.2012.02.038 n Corresponding author. Tel.: þ91 2026332783. E-mail address: vijayafmc@gmail.com (V. Chinnadurai). Pattern Recognition 45 (2012) 3501–3511