Optimisation of 2D U-Net Model Components for Automatic Prostate Segmentation on MRI

Abstract

In this paper, we develop an optimised state-of-the-art 2D U-Net model by studying the effects of the individual deep learning model components in performing prostate segmentation. We found that for upsampling, the combination of interpolation and convolution is better than the use of transposed convolution. For combining feature maps in each convolution block, it is only beneficial if a skip connection with concatenation is used. With respect to pooling, average pooling is better than strided-convolution, max, RMS or L2 pooling. Introducing a batch normalisation layer before the activation layer gives further performance improvement. The optimisation is based on a private dataset as it has a fixed 2D resolution and voxel size for every image which mitigates the need of a resizing operation in the data preparation process. Non-enhancing data preprocessing was applied and five-fold cross-validation was used to evaluate the fully automatic segmentation approach. We show it outperforms the traditional methods that were previously applied on the private dataset, as well as outperforming other comparable state-of-the-art 2D models on the public dataset PROMISE12.

Magnetic Resonance Imaging Based Radiomic Models of Prostate Cancer: A Narrative Review

“…Advanced deep learning algorithms have deployed convolutional neural network (CNN) to segment the ROI corresponding to a PCa [ 106 , 107 , 108 , 109 , 110 , 111 , 112 , 113 , 114 ]. The most common model used is the U-Net architecture, which is proposed for fully automatic segmentation of PCa with a DSC of ≥0.89.…”

Pyramidal Segmentation of Medical Images using Adversarial Training

“…This differs from detection, which would indicate the presence of a finding by classifying all pixels in the entire image as a whole. Several machine learning algorithms have provided state-of-the-art results for medical image segmentation [19][20][21][22][23][24], still improvements are needed to reach clinical usability, which motivates further research of machine learning-based segmentation approaches.…”

CDA-Net for Automatic Prostate Segmentation in MR Images

“…The comparison results indicate that the value of DSC of our algorithm on the PROMISE12 dataset is significantly higher than other popular algorithms, which means that the predicted segmentation result of our algorithm is closest to the real segmentation mask. [50] 89.00 Deep dense multi-path neural network [51] 89.01 Atlas registration and ensemble deep convolutional neural network [25] 91.00 HD-net [52] 91.35 nnU-Net [53] 91.61 BOWDA-Net [54] 92.54 CDA-Net (Proposed) 92.88…”