Showing papers by "Andrea Schenk published in 2018"

Automatic liver tumor segmentation in CT with fully convolutional neural networks and object-based postprocessing

Abstract: Automatic liver tumor segmentation would have a big impact on liver therapy planning procedures and follow-up assessment, thanks to standardization and incorporation of full volumetric information. In this work, we develop a fully automatic method for liver tumor segmentation in CT images based on a 2D fully convolutional neural network with an object-based postprocessing step. We describe our experiments on the LiTS challenge training data set and evaluate segmentation and detection performance. Our proposed design cascading two models working on voxel- and object-level allowed for a significant reduction of false positive findings by 85% when compared with the raw neural network output. In comparison with the human performance, our approach achieves a similar segmentation quality for detected tumors (mean Dice 0.69 vs. 0.72), but is inferior in the detection performance (recall 63% vs. 92%). Finally, we describe how we participated in the LiTS challenge and achieved state-of-the-art performance.

Deep learning based automatic liver tumor segmentation in CT with shape-based post-processing

Abstract: Accurate automatic liver tumor segmentation would have a big impact on liver therapy planning procedures and follow-up reporting, thanks to automation, standardization and incorporation of full volumetric information. In this work, we propose a fully automatic method for liver tumor segmentation in CT images based on a 2D convolutional deep neural network with a shape-based post-processing. We ran our experiments on the LiTS dataset and evaluated detection and segmentation performance. Our proposed method achieves segmentation quality for detected lesions comparable to a human expert and is able to detect 77% of potentially measurable tumor lesions according to the RECIST 1.1 guidelines. We submitted our results to the LiTS challenge achieving state-of-the-art performance.

Comparison of U-net-based Convolutional Neural Networks for Liver Segmentation in CT.

Abstract: Various approaches for liver segmentation in CT have been proposed: Besides statistical shape models, which played a major role in this research area, novel approaches on the basis of convolutional neural networks have been introduced recently. Using a set of 219 liver CT datasets with reference segmentations from liver surgery planning, we evaluate the performance of several neural network classifiers based on 2D and 3D U-net architectures. An interesting observation is that slice-wise approaches perform surprisingly well, with mean and median Dice coefficients above 0.97, and may be preferable over 3D approaches given current hardware and software limitations.

Focused scores enable reliable discrimination of small differences in steatosis.

Abstract: Automated image analysis enables quantitative measurement of steatosis in histological images. However, spatial heterogeneity of steatosis can make quantitative steatosis scores unreliable. To improve the reliability, we have developed novel scores that are “focused” on steatotic tissue areas. Focused scores use concepts of tile-based hotspot analysis in order to compute statistics about steatotic tissue areas in an objective way. We evaluated focused scores on three data sets of images of rodent liver sections exhibiting different amounts of dietary-induced steatosis. The same evaluation was conducted with the standard steatosis score computed by most image analysis methods. The standard score reliably discriminated large differences in steatosis (intraclass correlation coefficient ICC = 0.86), but failed to discriminate small (ICC = 0.54) and very small (ICC = 0.14) differences. With an appropriate tile size, mean-based focused scores reliably discriminated large (ICC = 0.92), small (ICC = 0.86) and very small (ICC = 0.83) differences. Focused scores based on high percentiles showed promise in further improving the discrimination of very small differences (ICC = 0.93). Focused scores enable reliable discrimination of small differences in steatosis in histological images. They are conceptually simple and straightforward to use in research studies.

Landmark-based evaluation of a deformable motion correction for DCE-MRI of the liver

Abstract: Annotation of meaningful landmark ground truth on DCE-MRI is difficult and laborious. Motion correction methods applied to DCE-MRI of the liver are thus mostly evaluated using qualitative or indirect measures. We propose a novel landmark annotation scheme that facilitates the generation of landmark ground truth on larger clinical datasets. In our annotation scheme, landmarks are equally distributed over all time points of all available dataset cases and annotated by multiple observers on a per-pair basis. The scheme is used to annotate 26 DCE-MRI of the liver. A subset of the ground truth is used to optimize parameters of a deformable motion correction. Several variants of the motion correction are evaluated on the remaining cases with respect to distances of corresponding landmarks after registration, deformation field properties, and qualitative measures. A landmark ground truth on 26 cases could be generated in under 12 h per observer with a mean inter-observer distance below the mean voxel diagonal. Furthermore, the landmarks are spatially well distributed within the liver. Parameter optimization significantly improves the performance of the motion correction, and landmark distance after registration is 2 mm. Qualitative evaluation of the motion correction reflects the quantitative results. The annotation scheme makes a landmark-based evaluation of motion corrections for hepatic DCE-MRI practically feasible for larger clinical datasets. The comparably large number of cases enables both optimization and evaluation of motion correction methods.

Younger Age is an Independent Factor for Graft Weight Overestimation: Analysis of the Clinical Impact on Recipient Outcomes in 340 Japanese Living Liver Donors

Abstract: Accurate preoperative estimation of graft weight is essential for improving outcomes in living donor liver transplantation. This retrospective study sought to identify factors associated with graft weight overestimation. From April 2006 to August 2015, 340 living donors were assigned to no-overestimate (n = 284) or overestimate (n = 56) groups. We defined graft weight overestimation as a discrepancy ≥15% between estimated graft volume and actual graft weight. Donor data were compared, and associated factors for graft weight overestimation were analyzed. Recipient outcomes were compared between the groups according to identified factors. Donors were significantly younger in the overestimate group than in the no-overestimate group (35.0 vs. 46.0 years; p < 0.001). Multivariate analysis identified donor age <45 years as an independent risk factor for graft weight overestimation (odds ratio 2.068; 95% confidence interval 1.114–3.839; p = 0.021). Among recipients with donors <45 years (n = 168), incidence of small-for-size dysfunction (SFSD) was significantly higher in the overestimate group than in the no-overestimate group (7/37 patients vs. 7/131 patients; p = 0.016); no significant difference was observed among recipients with donors ≥45 years (n = 172). First-year mortality was lower in SFSD recipients with donors <45 years (14.3 vs. 60.9%, p = 0.007). Among recipients with younger donors, graft survival was not significantly different between overestimate and no-overestimate groups. Younger donor age was an independent risk factor for graft weight overestimation leading to SFSD in recipients, but did not impair graft survival.