Liver cancer is one of the leading causes of cancer death. To assist doctors in hepatocellular carcinoma diagnosis and treatment planning, an accurate and automatic liver and tumor segmentation method is highly demanded in the clinical practice. Recently, fully convolutional neural networks (FCNs), including 2-D and 3-D FCNs, serve as the backbone in many volumetric image segmentation. However, 2-D convolutions cannot fully leverage the spatial information along the third dimension while 3-D convolutions suffer from high computational cost and GPU memory consumption. To address these issues, we propose a novel hybrid densely connected UNet (H-DenseUNet), which consists of a 2-D DenseUNet for efficiently extracting intra-slice features and a 3-D counterpart for hierarchically aggregating volumetric contexts under the spirit of the auto-context algorithm for liver and tumor segmentation. We formulate the learning process of the H-DenseUNet in an end-to-end manner, where the intra-slice representations and inter-slice features can be jointly optimized through a hybrid feature fusion layer. We extensively evaluated our method on the data set of the MICCAI 2017 Liver Tumor Segmentation Challenge and 3DIRCADb data set. Our method outperformed other state-of-the-arts on the segmentation results of tumors and achieved very competitive performance for liver segmentation even with a single model.

H-DenseUNet: Hybrid Densely Connected UNet for Liver and Tumor Segmentation From CT Volumes

The objective of this study was to develop anatomically correct whole body human models of an adult male (34 years old), an adult female (26 years old) and two children (an 11-year-old girl and a six-year-old boy) for the optimized evaluation of electromagnetic exposure. These four models are referred to as the Virtual Family. They are based on high resolution magnetic resonance (MR) images of healthy volunteers. More than 80 different tissue types were distinguished during the segmentation. To improve the accuracy and the effectiveness of the segmentation, a novel semi-automated tool was used to analyze and segment the data. All tissues and organs were reconstructed as three-dimensional (3D) unstructured triangulated surface objects, yielding high precision images of individual features of the body. This greatly enhances the meshing flexibility and the accuracy with respect to thin tissue layers and small organs in comparison with the traditional voxel-based representation of anatomical models. Conformal computational techniques were also applied. The techniques and tools developed in this study can be used to more effectively develop future models and further improve the accuracy of the models for various applications. For research purposes, the four models are provided for free to the scientific community.

/pdf/the-virtual-family-development-of-surface-based-anatomical-47u5t3e6am.pdf

The Virtual Family—development of surface-based anatomical models of two adults and two children for dosimetric simulations

The use of laparoscopy for liver surgery is increasing rapidly. The Second International Consensus Conference on Laparoscopic Liver Resections (LLR) was held in Morioka, Japan, from October 4 to 6, 2014 to evaluate the current status of laparoscopic liver surgery and to provide recommendations to aid its future development. Seventeen questions were addressed. The first 7 questions focused on outcomes that reflect the benefits and risks of LLR. These questions were addressed using the Zurich-Danish consensus conference model in which the literature and expert opinion were weighed by a 9-member jury, who evaluated LLR outcomes using GRADE and a list of comparators. The jury also graded LLRs by the Balliol Classification of IDEAL. The jury concluded that MINOR LLRs had become standard practice (IDEAL 3) and that MAJOR liver resections were still innovative procedures in the exploration phase (IDEAL 2b). Continued cautious introduction of MAJOR LLRs was recommended. All of the evidence available for scrutiny was of LOW quality by GRADE, which prompted the recommendation for higher quality evaluative studies. The last 10 questions focused on technical questions and the recommendations were based on literature review and expert panel opinion. Recommendations were made regarding preoperative evaluation, bleeding controls, transection methods, anatomic approaches, and equipment. Both experts and jury recognized the need for a formal structure of education for those interested in performing major laparoscopic LLR because of the steep learning curve.

Recommendations for laparoscopic liver resection: a report from the second international consensus conference held in Morioka

Efficient segmentation of globally optimal surfaces representing object boundaries in volumetric data sets is important and challenging in many medical image analysis applications. We have developed an optimal surface detection method capable of simultaneously detecting multiple interacting surfaces, in which the optimality is controlled by the cost functions designed for individual surfaces and by several geometric constraints defining the surface smoothness and interrelations. The method solves the surface segmentation problem by transforming it into computing a minimum s{\hbox{-}} t cut in a derived arc-weighted directed graph. The proposed algorithm has a low-order polynomial time complexity and is computationally efficient. It has been extensively validated on more than 300 computer-synthetic volumetric images, 72 CT-scanned data sets of different-sized plexiglas tubes, and tens of medical images spanning various imaging modalities. In all cases, the approach yielded highly accurate results. Our approach can be readily extended to higher-dimensional image segmentation.

Optimal Surface Segmentation in Volumetric Images-A Graph-Theoretic Approach

Learning Deconvolution Network for Semantic Segmentation

HepaVision2, a user friendly software application for preoperative planning based on CT images in liver surgery is presented. It is intended for both, evaluation of potential donors in living-related liver transplantation and planning of oncologic resections. The planning takes into account the patient’s individual anatomy allowing for fully automatic calculation of individual resection proposals including volumetric analysis. The results are visualized in 3D, thus allowing the surgeon to choose the optimal strategy for each patient. The software was tested in over 50 cases by our clinical partners and our institution. Average time needed per case is below one hour, therefore allowing the use of the software application in clinical routine.

HepaVision2 — a software assistant for preoperative planning in living-related liver transplantation and oncologic liver surgery

Preoperative volume prediction in adult living donor liver transplantation: how much can we rely on it?

Purpose 
To demonstrate and quantify the heat sink effect in hepatic microwave ablation (MWA) in a standardized ex vivo model, and to analyze the influence of vessel distance and blood flow on lesion volume and shape.


Materials and Methods 
108 ex vivo MWA procedures were performed in freshly harvested pig livers. Antennas were inserted parallel to non-perfused and perfused (700,1400 ml/min) glass tubes (diameter 5mm) at different distances (10, 15, 20mm). Ablation zones (radius, area) were analyzed and compared (Kruskal-Wallis Test, Dunn’s multiple comparison Test). Temperature changes adjacent to the tubes were measured throughout the ablation cycle.


Results 
Maximum temperature decreased significantly with increasing flow and distance (p<0.05). Compared to non-perfused tubes, ablation zones were significantly deformed by perfused tubes within 15mm distance to the antenna (p<0.05). At a flow rate of 700ml/min ablation zone radius was reduced to 37.2% and 80.1% at 10 and 15mm tube distance, respectively; ablation zone area was reduced to 50.5% and 89.7%, respectively.


Conclusion 
Significant changes of ablation zones were demonstrated in a pig liver model. Considerable heat sink effect was observed within a diameter of 15mm around simulated vessels, dependent on flow rate. This has to be taken into account when ablating liver lesions close to vessels.

/pdf/experimental-evaluation-of-the-heat-sink-effect-in-hepatic-3wuvhj8lsj.pdf

Experimental Evaluation of the Heat Sink Effect in Hepatic Microwave Ablation

Objectives Aim of the study was to evaluate the precise influence of different intrahepatic vessels, vessel sizes, and distances from the applicator on volume and shape of hepatic laser ablation zones in an in vivo porcine model. Materials and methods The study was approved by the institutional animal care and use committee. Eighteen computed tomography-guided Nd:YAG laser ablations were performed in the livers of 10 pigs at varying distances from hepatic veins and portal fields. After hepatectomy the livers were cut into 2-mm slices perpendicular to the laser applicator axes. For each ablation zone the maximum achievable (ideal) volume, the segmented (real) volume, the maximum radius, and the radius at the position of adjacent hepatic vessels were determined. The shapes of the ablation zones were evaluated qualitatively. Comparative statistics using the unpaired t test and a multiple linear regression analysis were performed. Results Ideal and real ablation zone volumes differed by 27.3% (8.6 +/- 1.5 mL vs. 6.4 +/- 1.1 mL; P 0.05). When influencing, all hepatic veins showed a focal indentation whereas portal fields always showed broad flattening of the ablation zone. Conclusions Portal fields lead to more heat sink than hepatic veins. The effects decreased with the distance between vessel and applicator tip, but less so for portal fields. The 2 vessel types induced considerably different shape alterations of the ablation zones. These results were not dependent on vessel size. This should be considered in the planning of thermal tumor ablations.

Influence of intrahepatic vessels on volume and shape of percutaneous thermal ablation zones: in vivo evaluation in a porcine model.

We present a fully automatic method employing convolutional neural networks based on the 2D U-net architecture and random forest classifier to solve the automatic liver lesion segmentation problem of the ISBI 2017 Liver Tumor Segmentation Challenge (LiTS). In order to constrain the ROI in which the tumors could be located, a liver segmentation is performed first. For the organ segmentation, an ensemble of convolutional networks is trained to segment a liver using a set of 179 liver CT datasets from liver surgery planning. Inside of the liver ROI a neural network, trained using 127 challenge training datasets, identifies tumor candidates, which are subsequently filtered with a random forest classifier yielding the final tumor segmentation. The evaluation on the 70 challenge test cases resulted in a mean Dice coefficient of 0.65, ranking our method in the second place.

Andrea Schenk

Papers

HepaVision2 — a software assistant for preoperative planning in living-related liver transplantation and oncologic liver surgery

Preoperative volume prediction in adult living donor liver transplantation: how much can we rely on it?

Experimental Evaluation of the Heat Sink Effect in Hepatic Microwave Ablation

Influence of intrahepatic vessels on volume and shape of percutaneous thermal ablation zones: in vivo evaluation in a porcine model.

Neural Network-Based Automatic Liver Tumor Segmentation With Random Forest-Based Candidate Filtering