Abstract
Brain tumor segmentation will accelerate the treatment process while diagnosing the patient's brain tumor size and life expectancy. Previously skilled doctors were used for segmenting the tumor region in brain MRI manually. Manually segmenting the brain tumor will take lots of manual work hours and also is error-prone. Recent advancements in computer vision and semantic segmentation made medical image segmentation more accurate and efficient. Improvements in medical image segmentation are being made in various aspects according to the data requirements. There are several different types of medical images, including MRI scans, CT-scans, X-Rays, and Ultrasound images. In this project, we use MRI scans of the brain. Detecting Brain tumors using MRI images is challenging because the images are low-contrasted generally, have tentacle-like structures, and are often diffused. One of the most challenging tasks during segmentation is preserving the edges. In this kind of medical images, the segmentation process usually needs a tremendous amount of memory and is time-consuming. To overcome these issues, we propose a framework to increase the image's quality (which helps us preserve every detail of the image). And an encoder-decoder based neural network model consists of octave convolution blocks, attentive batch normalization layers, and an active contour loss function. Our experiments showed that the proposed architecture performs better than many previously best-performed models by achieving better or comparable performance and substantially higher processing speed in segmenting brain tumors.