Automatic 3D liver location and segmentation via convolutional neural networks and graph cut (1605.03012v1)

Abstract: Purpose Segmentation of the liver from abdominal computed tomography (CT) image is an essential step in some computer assisted clinical interventions, such as surgery planning for living donor liver transplant (LDLT), radiotherapy and volume measurement. In this work, we develop a deep learning algorithm with graph cut refinement to automatically segment liver in CT scans. Methods The proposed method consists of two main steps: (i) simultaneously liver detection and probabilistic segmentation using 3D convolutional neural networks (CNNs); (ii) accuracy refinement of initial segmentation with graph cut and the previously learned probability map. Results The proposed approach was validated on forty CT volumes taken from two public databases MICCAI-Sliver07 and 3Dircadb. For the MICCAI-Sliver07 test set, the calculated mean ratios of volumetric overlap error (VOE), relative volume difference (RVD), average symmetric surface distance (ASD), root mean square symmetric surface distance (RMSD) and maximum symmetric surface distance (MSD) are 5.9%, 2.7%, 0.91%, 1.88 mm, and 18.94 mm, respectively. In the case of 20 3Dircadb data, the calculated mean ratios of VOE, RVD, ASD, RMSD and MSD are 9.36%, 0.97%, 1.89%, 4.15 mm and 33.14 mm, respectively. Conclusion The proposed method is fully automatic without any user interaction. Quantitative results reveal that the proposed approach is efficient and accurate for hepatic volume estimation in a clinical setup. The high correlation between the automatic and manual references shows that the proposed method can be good enough to replace the time-consuming and non-reproducible manual segmentation method.

• The paper presents a two-phase framework that first applies 3D CNNs for probabilistic liver detection and then refines the segmentation with graph cut optimization.
• Key performance metrics show a mean Volumetric Overlap Error of 5.9% and robust segmentation accuracy across multiple datasets.
• The approach streamlines liver segmentation in CT images, reducing manual effort and increasing reproducibility for critical clinical applications.

Automatic 3D Liver Location and Segmentation Using CNNs and Graph Cut

The paper presents a technique for the automatic segmentation of the liver in 3D computed tomography (CT) images using convolutional neural networks (CNNs) combined with graph cut optimization. The motivation for this work lies in the clinical importance of liver segmentation for surgical planning in interventions such as living donor liver transplants, radiotherapy, and volume measurement. Manual segmentation is known to be subjective, time-demanding, and lacking reproducibility, hence the necessity for an automated approach.

In this paper, the proposed method is partitioned into two primary phases: a liver detection and probabilistic segmentation using 3D CNNs, followed by refinement through graph cut using the learned probability maps. The CNNs are tasked with segmenting the liver by learning a hierarchy of features, an attribute that makes them suitable for tackling complex mappings inherent in image detection tasks.

In detail, the 3D CNNs proposed consist of several layers that extract features from the input CT images, gradually transitioning from low-level to high-level representations. The CNN architecture is explicitly designed to handle the 3D nature of the input data and to capture spatial hierarchies over volumetric inputs. Key performance metrics include the Volumetric Overlap Error (VOE), Relative Volume Difference (RVD), Average Symmetric Surface Distance (ASD), Root Mean Square Symmetric Surface Distance (RMSD), and Maximum Symmetric Surface Distance (MSD). Evaluation occurs over two datasets: MICCAI-Sliver07 and 3Dircabd, with results indicating a high correlation between automatic and manual segmentations.

Numerical results displayed a mean VOE of 5.9\% and RVD of 2.7\% for the MICCAI-Sliver07 test set, alongside an ASD of 0.91 mm, demonstrating improvement over many existing methodologies. Comparative analysis shows the method approximating or exceeding standard performance benchmarks in automatic liver segmentation. Notably, the system exhibits robustness across cases of normal and abnormal liver conditions, though it confronts issues with rotational invariance and similarity of liver and surrounding tissue intensities, suggesting areas for continued improvement.

The fusion of CNNs with graph cut methodologies allows for an improved capture of intensity inhomogeneities and anatomical variabilities, tackling common segmentation challenges related to surrounding organs' similar intensity distributions and varying liver shapes. This work significantly contributes by not only applying 3D CNNs to liver segmentation but also integrating them with graph cut to refine segmentation boundaries, optimizing overall performance and making it potentially applicable in a real clinical setup.

Future work could focus on enhancing the model's capability to manage edge cases involving similar intensity distributions and extend applicability to the segmentation of other organs. The paper underlines the momentum gathering behind incorporating machine learning techniques in medical image analysis, underscoring their potential in fulfilling complex diagnostic tasks and reinforcing their applicability in diverse medical contexts.