The process of segmenting medical images is incredibly significant for clinicians since it helps them deliver remedies that are both quick and reliable. There are numerous different ways, ranging from conventional techniques to convolutional neural network methods, for the identification of tumors in brain. The segmentation of the brain tumor by the standard approaches is successful; however, the image that results contains either noise or aberrations. This is the primary drawback of these methods. We discovered that max pooling layers have the ability to minimize the amount of aberration or noise that occurs during the process of image segmentation. We present an improved approach that employs the architecture of the deeplabv3+ and resnet50 as its encoder and additional max pooling layers in the bottle neck connections of the encoder with a pooling window of 5 by 5. The goal of this approach is to decrease the distortions that are shown in MRI images of brain tumors. When compared with a variety of conventional approaches, our technique has shown outstanding results.