MAD‐UNet: A deep U‐shaped network combined with an attention mechanism for pancreas segmentation in CT images

Abstract

Purpose Pancreas segmentation is a difficult task because of the high intrapatient variability in the shape, size, and location of the organ, as well as the low contrast and small footprint of the CT scan. At present, the U‐Net model is likely to lead to the problems of intraclass inconsistency and interclass indistinction in pancreas segmentation. To solve this problem, we improved the contextual and semantic feature information acquisition method of the biomedical image segmentation model (U‐Net) based on a convolutional network and proposed an improved segmentation model called the multiscale attention dense residual U‐shaped network (MAD‐UNet). Methods There are two aspects considered in this method. First, we adopted dense residual blocks and weighted binary cross‐entropy to enhance the semantic features to learn the details of the pancreas. Using such an approach can reduce the effects of intraclass inconsistency. Second, we used an attention mechanism and multiscale convolution to enrich the contextual information and suppress learning in unrelated areas. We let the model be more sensitive to pancreatic marginal information and reduced the impact of interclass indistinction. Results We evaluated our model using fourfold cross‐validation on 82 abdominal enhanced three‐dimensional (3D) CT scans from the National Institutes of Health (NIH‐82) and 281 3D CT scans from the 2018 MICCAI segmentation decathlon challenge (MSD). The experimental results showed that our method achieved state‐of‐the‐art performance on the two pancreatic datasets. The mean Dice coefficients were 86.10% ± 3.52% and 88.50% ± 3.70%. Conclusions Our model can effectively solve the problems of intraclass inconsistency and interclass indistinction in the segmentation of the pancreas, and it has value in clinical application. Code is available at https://github.com/Mrqins/pancreas-segmentation.

Artificial Intelligence Applied to Pancreatic Imaging: A Narrative Review

“…In the last few years, multiple CAD software approaches have been developed and tested on pancreatic imaging to improve the accuracy of examinations and the clinical decision-making process [ 62 , 63 , 64 , 65 , 66 , 67 , 68 , 69 , 70 , 71 , 72 , 73 , 74 , 75 , 76 , 77 ]. They have shown promising results for segmentation [ 65 , 66 , 67 , 68 , 69 , 70 , 71 , 72 , 77 ], tumor diagnosis, and classification [ 63 , 64 , 73 , 74 , 75 ] ( Figure 3 ).…”

“…The main application of DL algorithms for non-oncological studies was the automatic segmentation of pancreas and pancreatic lesions, which can support diagnosis and treatment planning and reduce the workload. [ 66 , 67 , 68 , 69 , 70 , 71 , 72 ] Gibson et al [ 66 ] used a deep learning-based segmentation algorithm for eight organs including the pancreas. Their model achieved a Dice similarity coefficient (DSC) of 0.78 vs. 0.71, 0.74, and 0.74.…”

MCMC Guided CNN Training and Segmentation for Pancreas Extraction

“…Another type of recent efforts focused on solving data scarcity and class imbalance issues through more generic learning techniques such as weighted loss function and data augmentation. For example, a weighted combination of DICE and Binary Cross Entropy loss functions was explored in [29] that improved overall learning capability and segmentation results. Multiple data augmentation strategies, including mixup [34] and RICAP [35] were studied in [36], leading to consistent improvement on pancreas segmentation for various U-Net architectures.…”

“…The work in [28] directly perform 3D learning with spatial attention, but they had to down-sample the input CT images first and use smaller batches for training. It is also difficult to derive a consistent bounding box based on prior knowledge due to large position variations [29].…”

“…Another way is to sample within a bounding box from approximate localization. However, in addition to the difficulty mentioned above [29], it is still unclear how to guarantee the consistency of patch distributions between training and testing stages.…”