Vanilla U-Net Architecture was a popular choice for segmentation in the domain of Medical Image Analysis and is used for automated diagnosis. While automated diagnosis has led to further enhancements in the capabilities of the Neural Network Models/Architectures it has also significantly increased the complexity of the network which also results in the high resource requirements. Given the complexity of medical images due to the texture and geometric orientations of the features, segmenting the features from medical images becomes a challenge. Addressing the above challenges, this article contributes to the field of medical image analysis. The paper exhibits the effectiveness of integrating the Atrous Spatial Pyramid Pooling (ASPP) block for increasing the segmenting capability of the basic U-Net architecture. This network model is then experimented with different activation functions, i.e., ReLU, Swish, Mish, and ELU to explore a solution that can address the vanishing gradient and dying neuron phenomenon. The results in the paper suggest that using the Mish activation function with the provided model architecture can outperform other activation functions, i.e., ReLU, Swish, and ELU.