Kersten Petersen

Deep Feature Learning for Knee Cartilage Segmentation Using a Triplanar Convolutional Neural Network

Segmentation of anatomical structures in medical images is often based on a voxel/pixel classification approach. Deep learning systems, such as convolutional neural networks (CNNs), can infer a hierarchical representation of images that fosters categorization. We propose a novel system for voxel classification integrating three 2D CNNs, which have a one-to-one association with the xy, yz and zx planes of 3D image, respectively. We applied our method to the segmentation of tibial cartilage in low field knee MRI scans and tested it on 114 unseen scans. Although our method uses only 2D features at a single scale, it performs better than a state-of-the-art method using 3D multi-scale features. In the latter approach, the features and the classifier have been carefully adapted to the problem at hand. That we were able to get better results by a deep learning architecture that autonomously learns the features from the images is the main insight of this study.

Unsupervised Deep Learning Applied to Breast Density Segmentation and Mammographic Risk Scoring

Mammographic risk scoring has commonly been automated by extracting a set of handcrafted features from mammograms, and relating the responses directly or indirectly to breast cancer risk. We present a method that learns a feature hierarchy from unlabeled data. When the learned features are used as the input to a simple classifier, two different tasks can be addressed: i) breast density segmentation, and ii) scoring of mammographic texture. The proposed model learns features at multiple scales. To control the models capacity a novel sparsity regularizer is introduced that incorporates both lifetime and population sparsity. We evaluated our method on three different clinical datasets. Our state-of-the-art results show that the learned breast density scores have a very strong positive relationship with manual ones, and that the learned texture scores are predictive of breast cancer. The model is easy to apply and generalizes to many other segmentation and scoring problems.

Breast Tissue Segmentation and Mammographic Risk Scoring Using Deep Learning

Mammographic scoring of density and texture are established methods to relate to the risk of breast cancer. We present a method that learns descriptive features from unlabeled mammograms and, using these learned features as the input to a simple classifier, address the following tasks: i) breast tissue segmentation ii) scoring of percentage mammographic density (PMD), and iii) scoring of mammographic texture (MT). Our results suggest that the learned PMD scores correlate well to manual ones, and that the learned MT scores are more related to future cancer risk than both manual and automatic PMD scores.

Mammographic density and structural features can individually and jointly contribute to breast cancer risk assessment in mammography screening: a case–control study

BackgroundMammographic density is a well-established risk factor for breast cancer. We investigated the association between three different methods of measuring density or parenchymal pattern/texture on digitized film-based mammograms, and examined to what extent textural features independently and jointly with density can improve the ability to identify screening women at increased risk of breast cancer.MethodsThe study included 121 cases and 259 age- and time matched controls based on a cohort of 14,736 women with negative screening mammograms from a population-based screening programme in Denmark in 2007 (followed until 31 December 2010). Mammograms were assessed using the Breast Imaging-Reporting and Data System (BI-RADS) density classification, Tabár’s classification on parenchymal patterns and a fully automated texture quantification technique. The individual and combined association with breast cancer was estimated using binary logistic regression to calculate Odds Ratios (ORs) and the area under the receiver operating characteristic (ROC) curves (AUCs).ResultsCases showed significantly higher BI-RADS and texture scores on average than controls (p < 0.001). All three methods were individually able to segregate women into different risk groups showing significant ORs for BI-RADS D3 and D4 (OR: 2.37; 1.32–4.25 and 3.93; 1.88–8.20), Tabár’s PIII and PIV (OR: 3.23; 1.20–8.75 and 4.40; 2.31–8.38), and the highest quartile of the texture score (3.04; 1.63–5.67). AUCs for BI-RADS, Tabár and the texture scores (continuous) were 0.63 (0.57–0–69), 0.65 (0.59–0–71) and 0.63 (0.57–0–69), respectively. Combining two or more methods increased model fit in all combinations, demonstrating the highest AUC of 0.69 (0.63-0.74) when all three methods were combined (a significant increase from standard BI-RADS alone).ConclusionOur findings suggest that the (relative) amount of fibroglandular tissue (density) and mammographic structural features (texture/parenchymal pattern) jointly can improve risk segregation of screening women, using information already available from normal screening routine, in respect to future personalized screening strategies.

A static SMC sampler on shapes for the automated segmentation of aortic calcifications

In this paper, we propose a sampling-based shape segmentation method that builds upon a global shape and a local appearance model. It is suited for challenging problems where there is high uncertainty about the correct solution due to a low signal-to-noise ratio, clutter, occlusions or an erroneous model. Our method suits for segmentation tasks where the number of objects is not known a priori, or where the object of interest is invisible and can only be inferred from other objects in the image. The method was inspired by shape particle filtering from de Bruijne and Nielsen, but shows substantial improvements to it. The principal contributions of this paper are as follows: (i) We introduce statistically motivated importance weights that lead to better performance and facilitate the application to new problems. (ii) We adapt the static sequential Monte Carlo (SMC) algorithm to the problem of image segmentation, where the algorithm proves to sample efficiently from high-dimensional static spaces. (iii) We evaluate the static SMC sampler on shapes on a medical problem of high relevance: the automated quantification of aortic calcifications on X-ray radiographs for the prognosis and diagnosis of cardiovascular disease and mortality. Our results suggest that the static SMC sampler on shapes is more generic, robust, and accurate than shape particle filtering, while being computationally equally costly.

A Bayesian Framework for Automated Cardiovascular Risk Scoring on Standard Lumbar Radiographs

We present a fully automated framework for scoring a patients risk of cardiovascular disease (CVD) and mortality from a standard lateral radiograph of the lumbar aorta. The framework segments abdominal aortic calcifications for computing a CVD risk score and performs a survival analysis to validate the score. Since the aorta is invisible on X-ray images, its position is reasoned from 1) the shape and location of the lumbar vertebrae and 2) the location, shape, and orientation of potential calcifications. The proposed framework follows the principle of Bayesian inference, which has several advantages in the complex task of segmenting aortic calcifications. Bayesian modeling allows us to compute CVD risk scores conditioned on the seen calcifications by formulating distributions, dependencies, and constraints on the unknown parameters. We evaluate the framework on two datasets consisting of 351 and 462 standard lumbar radiographs, respectively. Promising results indicate that the framework has potential applications in diagnosis, treatment planning, and the study of drug effects related to CVD.

Conditional point distribution models

In this paper, we propose an efficient method for drawing shape samples using a point distribution model (PDM) that is conditioned on given points. This technique is suited for sample-based segmentation methods that rely on a PDM, e.g. [6], [2] and [3]. It enables these algorithms to effectively constrain the solution space by considering a small number of user inputs -- often one or two landmarks are sufficient. The algorithm is easy to implement, highly efficient and usually converges in less than 10 iterations. We demonstrate how conditional PDMs based on a single user-specified vertebra landmark significantly improve the aorta and vertebrae segmentation on standard lateral radiographs. This is an important step towards a fast and cheap quantification of calcifications on X-ray radiographs for the prognosis and diagnosis of cardiovascular disease (CVD) and mortality

Automatic segmentation of vertebrae from radiographs: a sample-driven active shape model approach

Segmentation of vertebral contours is an essential task in the design of automatic tools for vertebral fracture assessment. In this paper, we propose a novel segmentation technique which does not require operator interaction. The proposed technique solves the segmentation problem in a hierarchical manner. In a first phase, a coarse estimate of the overall spine alignment and the vertebra locations is computed using a shape model sampling scheme. These samples are used to initialize a second phase of active shape model search, under a nonlinear model of vertebra appearance. The search is constrained by a conditional shape model, based on the variability of the coarse spine location estimates. The technique is evaluated on a data set of manually annotated lumbar radiographs. The results compare favorably to the previous work in automatic vertebra segmentation, in terms of both segmentation accuracy and failure rate.

89 GROWTH HORMONE ASSAYS: CLINICAL RESULTS IS OBTAINED WITH COMMERCIAL KITS

Much attention is being directed to the incidence of growth hormone deficiency and limits of HGH-concentration in respons to stimuli are under debate. In this study 23 serum samples from four patients were obtained during various stimulation test and analyzed by means of the following kits: Pharmacia (RIA 100), CIS (58-HGH-RIA equal to Sorin HGHK-2), Serono (HGH, RIA) and Hybritech (Tandem-R-HGH). All results were obtained in mU/l, the first three kits being standardized against WHO 66/217 and the last against HS 2243 E NIH. Pharmacia RIA-100 was the “in house method” (x) and the following correlations between the results obtained by other methods (y) were obtained:Samples of biosynthetic HGH: B-HGH was diluted to 22.1 mU/l and Pituitary standard P-HGH diluted to 18.8 mU/l, both from Nordisk Gentofte gave following results (mU/l):In conclusion methods based on one antibody-RIA (CIS=Sorin, Serono) gave the highest levels of HGH in plasma samples. Of the two IRMA-kits, Hybritech with two monoclonal antibodies gave the lowest results. Accordingly the dividing line between normal and partial deficiency is closely linked to the HGH assay chosen.

Normal ACTH levels in saltlosing congenital adrenal hyperplasia (CAH) with elevated plasma renin concentration (PRC). The effect of mineralocorticoid treatment

Eleven patients with salt losing CAH (21 OH deficiency), aged 5-19 years were treated with cortisone 20-55 mg/m2/24 h in 3 divided doses and supplementary salt. Normal (< 60 m.i.u./1) and moderately raised (133, 140 m.i.u./1) PRC values in 4 patients were compatible with good control - normal excretion of pregnanetriol (Ptriol). High PRC values (153-522 m.i.u./1) were found in connection with high values of Ptriol (6.5-34.8 μmol/24 h). Plasma Aldosterone (PA) was 3-11 (control values <18) ng/100 ml. Normal ACTH values (11-82 pg/ml) were found in all patients - in one pt. with high PRC (522 m.i.u./1) ACTH was 99 pg/ml.During maintenance of cortisone therapy, mineralocorticoid (Florinef(R) was added to the treatment. The dose was gradually increased over a period (median 15 months) up to 1-4 μg/kg/24 h (2 doses) and normal PRC values were established. PA decreased significantly to values below 4 ng/100 ml. Ptriol excretion decreased markedly (to 0.4 -69 μmol/24 h). ACTH values were still in the normal range. During the observation period no change in growth velocity could be demonstrated.