Brain Tumor Segmentation Using Gaussian-Based U-Net Architecture

As people’s age increases, they are more likely to get fatal because of the deadliest disease named brain tumor. A person’s survival rate of brain tumor is 30%, which is relatively low compared to other diseases. The brain tumor is caused because of the abnormal growth of cells around the brain which causes pressure on the healthy part of the brain and spreads around to other organs like liver, lungs, etc. They block the fluid around the brain and make it hard to survive. For medical diagnosis, it is one of the hectic tasks to find the tumor’s presence and position. So, We made an effort to make the segmentation which is accurate enough to predict the tumor type and its location. It makes the diagnosis much easier for medical professionals. Classes taken for segmentation are Eduma, Background, Enchancing, and Non-Enchancing. Several methods for segmenting brain tumors exist, but our team introduced a Gaussian-based lightweight U-Net architecture that yielded promising results. Gaussian smoothing preprocessing is used to cut the edges and blur the surrounding regions by reducing the noise which makes it suitable for segmentation of the BraTs2020 dataset. U-Net is a doubled architecture of the convolutional neural network model. The BraTS2020 dataset was utilized for our research, which has multi-modalities like T1-weighted MRI (T1), T1-weighted MRI with contrast (TCE), T2-weighted MRI (T2), segmented image, and Fluid-Attenuated Inversion Recovery (FLAIR). The accuracy, loss, and precision obtained by the research work are 0.9941, 0.0249, and 0.9927, respectively, which are comparatively better than other existing techniques.