Deep Learning approaches play a crucial role in medical imaging tasks aiding in accurate disease diagnosis and treatment planning. This paper presents ResUNet-AttnASPP, an improved ResUNet architecture incorporating Atrous Spatial Pyramid Pooling (ASPP) and attention gates for brain tumor segmentation from MRI data. ASPP detects multi-scale features, whereas attention gates tighten feature selection, increasing the segmentation efficiency. A comparison study is made between Unet, ReUnet and ResUNet-AttnASPP architectures using evaluation metrics like Intersection over Union (IoU) and Dice Similarity Coefficient. All these architectures are well suited to extract and process complex features from medical images. ResUnet advances UNet by adding residual connections, which enhance gradient flow and feature extraction abilities. ResUNet-AttnASPP outperforms U-Net and ResUNet, particularly for tumors with complex boundaries and geometries, resulting in higher segmentation accuracy and efficiency. Based on experimental results, ResUNet performs better than UNet in terms of segmentation accuracy, especially for tumors with intricate borders and geometries. The results demonstrate ResUNet-AttnASPP’s potential as a dependable architecture for automated brain tumor segmentation, hence boosting clinical diagnostics and medical imaging.