Cloud-Based Evaluation of Anatomical Structure Segmentation and Landmark Detection Algorithms: VISCERAL Anatomy Benchmarks.

Variations in the shape and appearance of anatomical structures in medical images are often relevant radiological signs of disease. Automatic tools can help automate parts of this manual process. A cloud-based evaluation framework is presented in this paper including results of benchmarking current state-of-the-art medical imaging algorithms for anatomical structure segmentation and landmark detection: the VISCERAL Anatomy benchmarks. The algorithms are implemented in virtual machines in the cloud where participants can only access the training data and can be run privately by the benchmark administrators to objectively compare their performance in an unseen common test set. Overall, 120 computed tomography and magnetic resonance patient volumes were manually annotated to create a standard Gold Corpus containing a total of 1295 structures and 1760 landmarks. Ten participants contributed with automatic algorithms for the organ segmentation task, and three for the landmark localization task. Different algorithms obtained the best scores in the four available imaging modalities and for subsets of anatomical structures. The annotation framework, resulting data set, evaluation setup, results and performance analysis from the three VISCERAL Anatomy benchmarks are presented in this article. Both the VISCERAL data set and Silver Corpus generated with the fusion of the participant algorithms on a larger set of non-manually-annotated medical images are available to the research community.

Comparison of 3D turbo spin-echo SPACE sequences with conventional 2D MRI sequences to assess the shoulder joint.

PURPOSE: To determine the accuracy and reliability of three-dimensional (3D) T1- and proton density (PD)-weighted turbo spin-echo (TSE) sampling perfection with application-optimized contrasts using different flip-angle evolution (SPACE) compared with conventional 2D sequences in assessment of the shoulder-joint. MATERIALS AND METHODS: Ninety-three subjects were examined on a 3-T MRI system with both conventional 2D-TSE sequences in T1-, T2- and PD-weighting and 3D SPACE sequences in T1- and PD-weighting. All examinations were assessed independently by two reviewers for common pathologies of the shoulder-joint. Agreement between 2D- and 3D-sequences and inter-observer-agreement was evaluated using kappa-statistics. RESULTS: Using conventional 2D TSE sequences as standard of reference, sensitivity, specificity, and accuracy values of 3D SPACE were 81.8%, 95.1%, and 93.5% for injuries of the supraspinatus-tendon (SSP), 81.3%, 93.5%, and 91.4% for the cartilage layer and 82.4%, 98.5%, and 97.5% for the long biceps tendon. Concordance between 2D and 3D was almost perfect for tendinopathies of the SSP (κ=0.85), osteoarthritis (κ=1), luxation of the biceps tendon (κ=1) and adjacent bone marrow (κ=0.92). Inter-observer-agreement was generally higher for conventional 2D TSE sequences (κ, 0.23-1.0), when compared to 3D SPACE sequences (κ, -0.33 to 1.0) except for disorders of the long biceps tendon and supraspinatus tendon rupture. CONCLUSION: Because of substantial and almost perfect concordance with conventional 2D TSE sequences for common shoulder pathologies, MRI examination-time can be reduced by nearly 40% (up to 11 min) using 3D-SPACE without loss of information.

Partially resected gliomas: diagnostic performance of fluid-attenuated inversion recovery MR imaging for detection of progression.

PURPOSE: To assess whether signal intensity (SI) different from that of cerebrospinal fluid (CSF) within the resection cavity during follow-up helps predict tumor progression in partially resected gliomas. MATERIALS AND METHODS: This retrospective study had local institutional review board approval, with waiver of informed consent. Seventy-five patients with partially resected and irradiated gliomas were evaluated. SI within the resection cavity on fluid-attenuated inversion recovery (FLAIR) magnetic resonance (MR) images was qualitatively and quantitatively assessed during follow-up. Qualitative analysis comprised visual comparison of SI in the resection cavity with SI of normal CSF by two readers. SI of the cavity was quantitatively assessed with region-of-interest (ROI) analysis normalized to background noise, contralateral healthy white matter, and CSF. Normalized SI during follow-up was compared with SI immediately after resection. Tumor progression was defined as increase in longest glioma diameter of at least 20% (Response Evaluation Criteria in Solid Tumors). Sensitivity and specificity of elevated SI in resection cavities for predicting or indicating tumor progression were calculated. Wilcoxon rank-sum test, Hodges-Lehman estimates, Kaplan-Meier curves, and linear mixed-effect models for repeated-measures data were used for quantitative SI measurements. RESULTS: Tumor progression at MR was seen in 44 patients (59%), and median progression-free survival was 4.1 years. Qualitative analysis showed that 25 of 44 patients with progression (57%) had SI increase in the resection cavity on FLAIR images. In 10 patients with progression (23%), SI increase was seen a mean of 5 months +/- 3 (standard deviation) before tumor size progression. In 15 patients with progression (34%), SI increase and tumor size progression were detected on the same MR study. In 19 patients with progressing glioma (43%), no SI increase was observed qualitatively. Among 31 patients without progression during follow-up (41%), no SI increase could be observed. Quantitative analysis showed no significant differences in ROI ratios at baseline (after surgery) between progressing and nonprogressing tumors, whereas significant differences in change of ROI ratios at the last measurement could be detected. Overall, SI increase on FLAIR images had specificity of 100% (95% confidence interval [CI]: 91%, 100%) and sensitivity of 57% (95% CI: 42%, 71%) for glioma progression. CONCLUSION: In partially resected gliomas, encapsulation of resection cavity, presumably by tumor cells, manifests as SI increase on FLAIR images and indicates tumor progression with very high specificity. (c) RSNA, 2010.