Brain Tumor Detection by Image Segmentation Using Customized UNet Deep Learning Based Model

Human brain is the most valuable organ that perform the most critical thinking and get the best solution methodology for real life problem. So, proper care should be taken to keep this valuable part be safe from being damaged by tumor disease. When a brain tumor is misdiagnosed, patients may receive the incorrect medical care, decreasing their chances of survival. Brain tumors are a deadly condition that, in its worst case, can have a very short life expectancy. In order to overcome these difficulties, the suggested framework uses CNN in large-scale trials to detect brain tumors utilizing the deep learning model's segmentation process. It is anticipated that the application of regularization strategies like augmentation and dropout will improve the precision of brain tumor identification with efficient manner. In this paper, we present a deep-learning method to detect brain tumors. We made use of a publicly available Kaggle brain tumor dataset that included color MRI pictures of both healthy and tumors brains that were afflicted. The dataset underwent preprocessing. A customized UNet CNN model was employed. Here, we customize the UNet model by adding 1 Convolution layer in downsampling and adding 1 De-Convolution layer in upsampling. With our suggested model, we achieved 99.80% train accuracy. For the validation and test phase, we achieved 99.78% & 99.75% accuracy, respectively.