Comparison of Various Colorization Techniques for MRI Brain Tumor Segmentation using Convolutional Neural Networks

In the realm of image processing and computer vision, image segmentation takes on a foundational role, serving as a critical inception task for diverse analyses, especially in the application of brain tumor detection. Given the grayscale nature of MRI brain tumor images, the primary objective of this study was to explore the effects of employing various colorization techniques on the accuracy of MRI brain tumor segmentation. The incorporation of color in images has the added benefit of enhancing edge delineation by approximately 10% compared with the grayscale version. To achieve this, a spectrum of color mappings including red, green, blue, yellow, HSV, pseudo-color, and deoldify was applied and compared with the original grayscale MRI images in terms of tumor segmentation accuracy. This comparative segmentation accuracy was performed using two well-established convolutional neural networks (CNNs), namely U-net and SegNet models. Experimental outcomes highlighted the green color mapping as the optimal selection for achieving effective segmentation of meningioma tumors in MRI brain images, followed by the grayscale, HSV, deoldify, pseudo-color, blue, red and yellow color maps. This conclusion was drawn due to the remarkable performance exhibited by the green color mapping across five evaluation metrics, attaining the highest overall average performance measures spanning Dice (0.8876), Jaccard (0.8897), Matthews correlation coefficient (0.8897), Accuracy (0.9954) and remarkably, the lowest recorded Error score (0.0046).