Early detection of the Autosomal Dominant Polycystic Kidney Disease (ADPKD) is crucial as it is one of the most common causes of end-stage renal disease (ESRD) that leads to kidney failure. Therefore, it is important to assess the disease progression to plan for proper therapeutic intervention. The total kidney volume (TKV) has been shown to increase with ADPKD progression and therefore can be used to quantify disease progression. However, TKV calculation requires accurate delineation of the kidney volumes, usually performed manually by an expert physician. Time-consuming manual contouring is a limitation for deploying deep learning medical image processing. Therefore, large annotated datasets are rare. In this work, we address this problem by implementing three attention mechanisms into the U-Net. In addition, we also implement a cosine loss function, that has been shown to work well on small datasets. Our results show significant improvement (p-value < 0.05) over the reference kidney segmentation U-Net. We show that the attention mechanisms and/or the cosine loss can help improve the dice score up to 91% (approx. 2-3% improvement) with a mean symmetric surface distance of 1.36 mm (11.2 % improvement) while utilizing in total only 100 datasets for training and testing.