Glioblastoma is one of the most aggressive tumors affecting adults. Magnetic resonance imaging (MRI) is widely used for diagnosis and monitoring disease status following intervention including craniotomy, irradiation and various chemotherapies. The Macdonald or Response Assessment in Neuro-Oncology (RANO) criteria based on tumor volume estimations using MRI are routinely used to determine tumor response to treatment. However, these methods are manual and prone to operator bias. They also do not consider tumor heterogeneity. Automated estimations of volumes of tumor sub-regions are necessary to follow disease status objectively. In this study, we developed and implemented and automated technique using morphological methodologies to identify and classify tumor sub-regions by integrating information from multi-modal MRI.Nine subjects (5 males, 4 females) were included in the study. Average age was 52.43 (± 11.87) years ranging from 29 to 66 years. Three-dimensional (1 mm3 isotropic voxel) T2-weighted (T2 images; TR/TE = 2500 ms/388.3 ms), fluid attenuated inversion recovery (FLAIR; TR/TE/TI = 4800 ms/323 ms/1650 ms), and T1-weighted (T1) pre- and post-contrast images (TR/TE = 8.1 ms/3.7 ms) were acquired. FLAIR, T1 pre-/post-contrast images were co-aligned with T2 images using rigid body registration technique [1]. The T2 images were skull-stripped [2] which was then used to mask the FLAIR and T1 images. The images were further processed for intensity non-uniformities correction and noise reduction [1]. Morphological grayscale reconstruction (MGR) by geodesic dilation was applied to FLAIR and T2 images to obtain regional maxima [3, 5-6]. The original image was first decomposed by the application of grayscale erosion with a three-dimensional structuring element (sphere) to obtain a modified image with voxel intensities lower than or equal to the intensities of corresponding voxel on original image. An elementary geodesic dilation was applied to the modified image by first dilating the image with the structuring element, followed by the calculation of point-wise minimum with the modified image resulting in the reconstructed image. The elementary geodesic dilation was applied iteratively until no further change in the reconstructed image was observed. Subtraction of reconstructed image from the original image resulted in regional maxima. The procedure identified all regional maxima in the image and therefore the application of it to FLAIR images results to tumor/vasogenic edema. But application of this procedure toT2 images identifies both cerebrospinal fluid (CSF) and tumor regions. Similarly, application of MGR by iterative application of elementary geodesic erosion to FLAIR image results in the identification of regional minima representing CSF [4, 5-6]. Therefore, tumor/vasogenic edema is extracted by combining the results of MGR on FLAIR and T2 images. On the other hand, application of MGR by geodesic dilation applied to T1 post-contrast images results in the identification of regional maxima that include enhanced part of the tumor and false classification representing vasculatures [2, 4-5]. The false classifications are eliminated by masking the results with the ratio maps (T1 pre-contrast - T1 post-contrast)/ T1 pre-contrast images. Our technique performed consistently well across all the patients. The raters who included a neuroradiologist and a neurooncologist judged the performance of this technique to be highly satisfactory in segmenting various sub-components of the tumor.Figure 1 shows the application of MGR to determine the regional maxima on FLAIR and T2 images at different iteration levels. In this example, the reconstructed image converged after 24 iterations. As can be observed from this figure, regional maxima consist of tumor region only on FLAIR, but include CSF on T2 image. Figure 2 shows an example of multi-channel images and the extracted tumor on FLAIR and enhanced part of tumor on the T1 post-contrast images. The neuroradiologist and neurooncologist felt that the automated program performed exceedingly well. We have developed, implemented, and validated an automated technique for identifying tumor and its sub-components with minimal false classification. This technique also appears to be very robust based on the consistent results obtained across all the patients. We expect this technique to play an important role in the management of glioblastoma patients.