Explainable Anatomical Shape Analysis Through Deep Hierarchical Generative Models.

Quantification of anatomical shape changes currently relies on scalar global indexes which are largely insensitive to regional or asymmetric modifications. Accurate assessment of pathology-driven anatomical remodeling is a crucial step for the diagnosis and treatment of many conditions. Deep learning approaches have recently achieved wide success in the analysis of medical images, but they lack interpretability in the feature extraction and decision processes. In this work, we propose a new interpretable deep learning model for shape analysis. In particular, we exploit deep generative networks to model a population of anatomical segmentations through a hierarchy of conditional latent variables. At the highest level of this hierarchy, a two-dimensional latent space is simultaneously optimised to discriminate distinct clinical conditions, enabling the direct visualisation of the classification space. Moreover, the anatomical variability encoded by this discriminative latent space can be visualised in the segmentation space thanks to the generative properties of the model, making the classification task transparent. This approach yielded high accuracy in the categorisation of healthy and remodelled left ventricles when tested on unseen segmentations from our own multi-centre dataset as well as in an external validation set, and on hippocampi from healthy controls and patients with Alzheimer's disease when tested on ADNI data. More importantly, it enabled the visualisation in three-dimensions of both global and regional anatomical features which better discriminate between the conditions under exam. The proposed approach scales effectively to large populations, facilitating high-throughput analysis of normal anatomy and pathology in large-scale studies of volumetric imaging.

Sex and regional differences in myocardial plasticity in aortic stenosis are revealed by 3D model machine learning.

AIMS: Left ventricular hypertrophy (LVH) in aortic stenosis (AS) varies widely before and after aortic valve replacement (AVR), and deeper phenotyping beyond traditional global measures may improve risk stratification. We hypothesized that machine learning derived 3D LV models may provide a more sensitive assessment of remodelling and sex-related differences in AS than conventional measurements. METHODS AND RESULTS: One hundred and sixteen patients with severe, symptomatic AS (54% male, 70 ± 10 years) underwent cardiovascular magnetic resonance pre-AVR and 1 year post-AVR. Computational analysis produced co-registered 3D models of wall thickness, which were compared with 40 propensity-matched healthy controls. Preoperative regional wall thickness and post-operative percentage wall thickness regression were analysed, stratified by sex. AS hypertrophy and regression post-AVR was non-uniform-greatest in the septum with more pronounced changes in males than females (wall thickness regression: -13 ± 3.6 vs. -6 ± 1.9%, respectively, P < 0.05). Even patients without LVH (16% with normal indexed LV mass, 79% female) had greater septal and inferior wall thickness compared with controls (8.8 ± 1.6 vs. 6.6 ± 1.2 mm, P < 0.05), which regressed post-AVR. These differences were not detectable by global measures of remodelling. Changes to clinical parameters post-AVR were also greater in males: N-terminal pro-brain natriuretic peptide (NT-proBNP) [-37 (interquartile range -88 to -2) vs. -1 (-24 to 11) ng/L, P = 0.008], and systolic blood pressure (12.9 ± 23 vs. 2.1 ± 17 mmHg, P = 0.009), with changes in NT-proBNP correlating with percentage LV mass regression in males only (ß 0.32, P = 0.02). CONCLUSION: In patients with severe AS, including those without overt LVH, LV remodelling is most plastic in the septum, and greater in males, both pre-AVR and post-AVR. Three-dimensional machine learning is more sensitive than conventional analysis to these changes, potentially enhancing risk stratification. CLINICAL TRIAL REGISTRATION: Regression of myocardial fibrosis after aortic valve replacement (RELIEF-AS); NCT02174471. https://clinicaltrials.gov/ct2/show/NCT02174471.

Primary Rectal Cancer: Repeatability of Global and Local-Regional MR Imaging Texture Features.

Purpose To assess the day-to-day repeatability of global and local-regional magnetic resonance (MR) imaging texture features derived from primary rectal cancer. Materials and Methods After ethical approval and patient informed consent were obtained, two pretreatment T2-weighted axial MR imaging studies performed prospectively with the same imaging unit on 2 consecutive days in 14 patients with rectal cancer (11 men [mean age, 61.7 years], three women [mean age, 70.0 years]) were analyzed to extract (a) global first-order statistical histogram and model-based fractal features reflecting the whole-tumor voxel intensity histogram distribution and repeating patterns, respectively, without spatial information and (b) local-regional second-order and high-order statistical texture features reflecting the intensity and spatial interrelationships between adjacent in-plane or multiplanar voxels or regions, respectively. Repeatability was assessed for 46 texture features, and mean difference, 95% limits of agreement, within-subject coefficient of variation (wCV), and repeatability coefficient (r) were recorded. Results Repeatability was better for global parameters than for most local-regional parameters. In particular, histogram mean, median, and entropy, fractal dimension mean and standard deviation, and second-order entropy, homogeneity, difference entropy, and inverse difference moment demonstrated good repeatability, with narrow limits of agreement and wCVs of 10% or lower. Repeatability was poorest for the following high-order gray-level run-length (GLRL) gray-level zone size matrix (GLZSM) and neighborhood gray-tone difference matrix (NGTDM) parameters: GLRL intensity variability, GLZSM short-zone emphasis, GLZSM intensity nonuniformity, GLZSM intensity variability, GLZSM size zone variability, and NGTDM complexity, demonstrating wider agreement limits and wCVs of 50% or greater. Conclusion MR imaging repeatability is better for global texture parameters than for local-regional texture parameters, indicating that global texture parameters should be sufficiently robust for clinical practice. Online supplemental material is available for this article.

The precision of textural analysis in (18)F-FDG-PET scans of oesophageal cancer.

OBJECTIVES: Measuring tumour heterogeneity by textural analysis in (18)F-fluorodeoxyglucose positron emission tomography ((18)F-FDG PET) provides predictive and prognostic information but technical aspects of image processing can influence parameter measurements. We therefore tested effects of image smoothing, segmentation and quantisation on the precision of heterogeneity measurements. METHODS: Sixty-four (18)F-FDG PET/CT images of oesophageal cancer were processed using different Gaussian smoothing levels (2.0, 2.5, 3.0, 3.5, 4.0 mm), maximum standardised uptake value (SUVmax) segmentation thresholds (45%, 50%, 55%, 60%) and quantisation (8, 16, 32, 64, 128 bin widths). Heterogeneity parameters included grey-level co-occurrence matrix (GLCM), grey-level run length matrix (GLRL), neighbourhood grey-tone difference matrix (NGTDM), grey-level size zone matrix (GLSZM) and fractal analysis methods. The concordance correlation coefficient (CCC) for the three processing variables was calculated for each heterogeneity parameter. RESULTS: Most parameters showed poor agreement between different bin widths (CCC median 0.08, range 0.004-0.99). Segmentation and smoothing showed smaller effects on precision (segmentation: CCC median 0.82, range 0.33-0.97; smoothing: CCC median 0.99, range 0.58-0.99). CONCLUSIONS: Smoothing and segmentation have only a small effect on the precision of heterogeneity measurements in (18)F-FDG PET data. However, quantisation often has larger effects, highlighting a need for further evaluation and standardisation of parameters for multicentre studies. KEY POINTS: • Heterogeneity measurement precision in (18) F-FDG PET is influenced by image processing methods. • Quantisation shows large effects on precision of heterogeneity parameters in (18) F-FDG PET/CT. • Smoothing and segmentation show comparatively smaller effects on precision of heterogeneity parameters.