Assessing the performance of deep learning-based models for prostate cancer segmentation using uncertainty scores