Abdominal multi-organ segmentation with organ-attention networks and statistical fusion (1804.08414v1)

Abstract: Accurate and robust segmentation of abdominal organs on CT is essential for many clinical applications such as computer-aided diagnosis and computer-aided surgery. But this task is challenging due to the weak boundaries of organs, the complexity of the background, and the variable sizes of different organs. To address these challenges, we introduce a novel framework for multi-organ segmentation by using organ-attention networks with reverse connections (OAN-RCs) which are applied to 2D views, of the 3D CT volume, and output estimates which are combined by statistical fusion exploiting structural similarity. OAN is a two-stage deep convolutional network, where deep network features from the first stage are combined with the original image, in a second stage, to reduce the complex background and enhance the discriminative information for the target organs. RCs are added to the first stage to give the lower layers semantic information thereby enabling them to adapt to the sizes of different organs. Our networks are trained on 2D views enabling us to use holistic information and allowing efficient computation. To compensate for the limited cross-sectional information of the original 3D volumetric CT, multi-sectional images are reconstructed from the three different 2D view directions. Then we combine the segmentation results from the different views using statistical fusion, with a novel term relating the structural similarity of the 2D views to the original 3D structure. To train the network and evaluate results, 13 structures were manually annotated by four human raters and confirmed by a senior expert on 236 normal cases. We tested our algorithm and computed Dice-Sorensen similarity coefficients and surface distances for evaluating our estimates of the 13 structures. Our experiments show that the proposed approach outperforms 2D- and 3D-patch based state-of-the-art methods.

• The paper presents a novel two-stage CNN using organ-attention networks and reverse connections to accurately segment abdominal organs in CT images.
• It integrates statistical fusion based on structural similarity to reconcile multiple 2D views into coherent 3D segmentation.
• The approach outperforms existing methods by improving Dice-Sørensen coefficients and reducing mean surface distances, especially for complex organs.

An Analysis of Abdominal Multi-organ Segmentation Using Organ-Attention Networks

This paper introduces a novel approach to abdominal organ segmentation from computed tomography (CT) images using Organ-Attention Networks with Reverse Connections (OAN-RC) integrated with a statistical fusion method based on structural similarity. The primary aim is to address the inherent challenges of organ segmentation in CT images, which include weak organ boundaries, the variable sizes of organs, and the complex surrounding tissue.

Methodology

The proposed technique enhances segmentation through a combination of two key innovations: organ-attention networks and structural similarity-based statistical fusion. The organ-attention network is designed as a two-stage convolutional neural network (CNN) that leverages an attention mechanism to reduce background noise and emphasize target organs. This effectively aids in focusing on the organ region with minimal distraction from adjacent structures. Reverse connections further enhance this network by providing lower layers with improved semantic information, thereby facilitating more accurate segmentation of organs irrespective of their sizes.

The authors train the network using two-dimensional (2D) views of the CT images, reconciling the computational limitations of three-dimensional (3D) deep networks while maximizing the use of image data. This approach allows the aggregation of holistic information across slices, enhancing the network's capacity to handle complex segmentation tasks.

Furthermore, the integration of segmentation outputs through statistical fusion exploits structural similarity between 2D views and the original 3D architecture of the CT data. This step serves to reconcile the information from various 2D planes and align it structurally in 3D space, thereby improving segmentation accuracy and consistency.

Results

The paper leverages 236 abdominal CT scans with manually annotated organ structures as a benchmark, utilizing four-fold cross-validation to test the model’s efficacy. Notably, the authors report that their approach yields superior performance compared to existing methods, particularly in terms of Dice-Sørensen similarity coefficients (DSC) and mean surface distances. For instance, segmentation accuracy showed significant improvements for complex abdominal organs like the pancreas and duodenum where clear boundary demarcations are typically challenging. These outcomes underscore the proposed method’s ability to better delineate organ structures even in complex anatomical arrangements.

Implications and Future Directions

The methodological innovations presented in this paper, namely the OAN-RC framework and statistical fusion strategy, offer significant contributions to medical image analysis. Practical applications range from aiding radiologists in diagnosis and treatment planning to enhancing the precision of computer-aided intervention systems. The efficient computational requirements of the system also make it viable for use in real-time diagnostic settings.

The implications of this research extend beyond immediate clinical applications. By demonstrating effective integration of attention mechanisms with statistical fusion, this work sets the stage for enhanced multi-organ segmentation in other body regions or imaging modalities. Future research directions could explore adaptation of this framework to other types of imaging data, such as MRI, or incorporating additional contextual information through integration with other AI-based diagnostic tools.

Conclusion

This paper successfully articulates an advanced method for multi-organ segmentation in medical imaging. By integrating a two-stage organ-attention network with reverse connections and a sophisticated statistical fusion algorithm, this research presents a powerful solution to the complexities of abdominal organ segmentation. This approach not only marks an advancement in addressing segmentation challenges but also lays a foundation for further exploration and enhancement of AI-driven medical imaging technologies.