Automated analysis of 3D-echocardiography using spatially registered patient-specific CMR meshes

Abstract Funding Acknowledgements Type of funding sources: Public grant(s) – National budget only. Main funding source(s): National Heart Foundation (NHF) of New Zealand Health Research Council (HRC) of New Zealand Artificial intelligence shows considerable promise for automated analysis and interpretation of medical images, particularly in the domain of cardiovascular imaging. While application to cardiac magnetic resonance (CMR) has demonstrated excellent results, automated analysis of 3D echocardiography (3D-echo) remains challenging, due to the lower signal-to-noise ratio (SNR), signal dropout, and greater interobserver variability in manual annotations. As 3D-echo is becoming increasingly widespread, robust analysis methods will substantially benefit patient evaluation. We sought to leverage the high SNR of CMR to provide training data for a convolutional neural network (CNN) capable of analysing 3D-echo. We imaged 73 participants (53 healthy volunteers, 20 patients with non-ischaemic cardiac disease) under both CMR and 3D-echo (<1 hour between scans). 3D models of the left ventricle (LV) were independently constructed from CMR and 3D-echo, and used to spatially align the image volumes using least squares fitting to a cardiac template. The resultant transformation was used to map the CMR mesh to the 3D-echo image. Alignment of mesh and image was verified through volume slicing and visual inspection (Fig. 1) for 120 paired datasets (including 47 rescans) each at end-diastole and end-systole. 100 datasets (80 for training, 20 for validation) were used to train a shallow CNN for mesh extraction from 3D-echo, optimised with a composite loss function consisting of normalised Euclidian distance (for 290 mesh points) and volume. Data augmentation was applied in the form of rotations and tilts (<15 degrees) about the long axis. The network was tested on the remaining 20 datasets (different participants) of varying image quality (Tab. I). For comparison, corresponding LV measurements from conventional manual analysis of 3D-echo and associated interobserver variability (for two observers) were also estimated. Initial results indicate that the use of embedded CMR meshes as training data for 3D-echo analysis is a promising alternative to manual analysis, with improved accuracy and precision compared with conventional methods. Further optimisations and a larger dataset are expected to improve network performance. (n = 20) LV EDV (ml) LV ESV (ml) LV EF (%) LV mass (g) Ground truth CMR 150.5 ± 29.5 57.9 ± 12.7 61.5 ± 3.4 128.1 ± 29.8 Algorithm error -13.3 ± 15.7 -1.4 ± 7.6 -2.8 ± 5.5 0.1 ± 20.9 Manual error -30.1 ± 21.0 -15.1 ± 12.4 3.0 ± 5.0 Not available Interobserver error 19.1 ± 14.3 14.4 ± 7.6 -6.4 ± 4.8 Not available Tab. 1. LV mass and volume differences (means ± standard deviations) for 20 test cases. Algorithm: CNN – CMR (as ground truth). Abstract Figure. Fig 1. CMR mesh registered to 3D-echo.

Download Full-text

Towards improving the accuracy of aortic transvalvular pressure gradients: rethinking Bernoulli

Medical & Biological Engineering & Computing ◽

10.1007/s11517-020-02186-w ◽

2020 ◽

Vol 58 (8) ◽

pp. 1667-1679

Author(s):

Benedikt Franke ◽

J. Weese ◽

I. Waechter-Stehle ◽

J. Brüning ◽

T. Kuehne ◽

...

Keyword(s):

Test Data ◽

Ground Truth ◽

Training Data ◽

Patient Specific ◽

Pressure Gradients ◽

Bernoulli Model ◽

Bernoulli Equation ◽

Data Set ◽

Non Invasive ◽

Adjusted Model

Abstract The transvalvular pressure gradient (TPG) is commonly estimated using the Bernoulli equation. However, the method is known to be inaccurate. Therefore, an adjusted Bernoulli model for accurate TPG assessment was developed and evaluated. Numerical simulations were used to calculate TPGCFD in patient-specific geometries of aortic stenosis as ground truth. Geometries, aortic valve areas (AVA), and flow rates were derived from computed tomography scans. Simulations were divided in a training data set (135 cases) and a test data set (36 cases). The training data was used to fit an adjusted Bernoulli model as a function of AVA and flow rate. The model-predicted TPGModel was evaluated using the test data set and also compared against the common Bernoulli equation (TPGB). TPGB and TPGModel both correlated well with TPGCFD (r > 0.94), but significantly overestimated it. The average difference between TPGModel and TPGCFD was much lower: 3.3 mmHg vs. 17.3 mmHg between TPGB and TPGCFD. Also, the standard error of estimate was lower for the adjusted model: SEEModel = 5.3 mmHg vs. SEEB = 22.3 mmHg. The adjusted model’s performance was more accurate than that of the conventional Bernoulli equation. The model might help to improve non-invasive assessment of TPG.

Download Full-text

Predicting Knee Adduction Moment Response to Gait Retraining with Minimal Clinical Data

10.1101/2021.09.29.462292 ◽

2021 ◽

Author(s):

Nataliya Rokhmanova ◽

Katherine J. Kuchenbecker ◽

Peter B. Shull ◽

Reed Ferber ◽

Eni Halilaj

Keyword(s):

Clinical Data ◽

Synthetic Data ◽

Ground Truth ◽

Absolute Error ◽

Training Data ◽

Knee Adduction Moment ◽

Patient Specific ◽

Gait Retraining ◽

Foot Progression Angle ◽

Public Datasets

Knee osteoarthritis is a progressive disease mediated by high joint loads. Foot progression angle modifications that reduce the knee adduction moment (KAM), a surrogate of knee loading, have demonstrated efficacy in alleviating pain and improving function. Although changes to the foot progression angle are overall beneficial, KAM reductions are not consistent across patients. Moreover, customized interventions are time-consuming and require instrumentation not commonly available in the clinic. We present a model that uses minimal clinical data to predict the extent of first peak KAM reduction after toe-in gait retraining. For such a model to generalize, the training data must be large and variable. Given the lack of large public datasets that contain different gaits for the same patient, we generated this dataset synthetically. Insights learned from ground-truth datasets with both baseline and toe-in gait trials (N=12) enabled the creation of a large (N=138) synthetic dataset for training the predictive model. On a test set of data collected by a separate research group (N=15), the first peak KAM reduction was predicted with a mean absolute error of 0.134% body weight * height (%BW*HT). This error is smaller than the test set’s subject average standard deviation of the first peak during baseline walking (0.306%BW*HT). This work demonstrates the feasibility of training predictive models with synthetic data and may provide clinicians with a streamlined pathway to identify a patient-specific gait retraining outcome without requiring gait lab instrumentation.

Download Full-text

P4360Personalised shape models of the left ventricle from 3D echocardiography: an initial comparison to cardiac magnetic resonance imaging

European Heart Journal ◽

10.1093/eurheartj/ehz745.0767 ◽

2019 ◽

Vol 40 (Supplement_1) ◽

Author(s):

D Zhao ◽

K Gilbert ◽

C R McDougal ◽

V Y Wang ◽

H Houle ◽

...

Keyword(s):

Magnetic Resonance Imaging ◽

Cardiac Magnetic Resonance ◽

Magnetic Resonance ◽

Cardiac Magnetic Resonance Imaging ◽

Left Ventricular ◽

Patient Specific ◽

Resonance Imaging ◽

Shape Models ◽

Lv Mass ◽

3D Echo

Abstract Background The heart constantly adapts to maintain cardiac output. In the longer term, this process (remodeling) can manifest as changes in ventricular volume, sphericity, and/or wall thickness, amongst several other morphological indices. Previous studies have shown the significance of remodeling in evaluations of survival, and as a determinant of the clinical course of heart failure. Yet surprisingly, diagnostic measures, typically of left ventricular (LV) mass and ejection fraction, neglect much of the shape information that is available through imaging. A recent UK Biobank study revealed that morphometric atlases show more compelling associations with cardiovascular risk factors, than do LV mass and volumes. While it has been possible to construct shape models from cardiac magnetic resonance imaging (MRI), such a framework is still under development for echocardiography (echo). Purpose Despite MRI being long regarded as the gold standard, it is greatly limited by high costs, long scan times and incompatibility with ferromagnetic cardiac devices. In contrast, echo has presented as a convenient alternative, whilst also offering good temporal resolution. The advancements of 3D echo now provide adequate spatial resolution and thus elicit the possibility of conducting more complex analyses on this modality. With the ability to extract LV geometry directly from 3D echo acquisitions, we sought to create dynamic, 3D patient-specific models–and subsequently compare these results to those derived from MRI. Methods As part of an ongoing study, 8 volunteers with no known cardiovascular problems (nor family history thereof), were recruited for non-invasive imaging. Cine MRI and 3D echo of the LV were acquired within a 2 hour session. A Siemens Avanto Fit 1.5 T MRI scanner and Siemens ACUSON SC2000 Ultrasound System with a 4Z1c Transducer were used. 3D models of the LV were generated independently from echo (EchobuildR 2.7 prototype software, Siemens Ultrasound) and MRI acquisitions (Cardiac Image Modeller v8.1), and registered to fiducial landmarks (apex, base plane, right ventricular inserts) and myocardial contours. Results Euclidian distances between 1682 corresponding points sampled from the surface of echo/MRI models were calculated, and used as a discrepancy measure (Figure). Across the 8 cases, we found an average root mean square deviation (RMSD) of 5.71 mm at end-systole and 6.03 mm at end-diastole. The maximum RMSD for a single model was 9.47 mm (case 8, ES). Conclusion We demonstrate that it is possible to create shape models from 3D echo examinations for comparison with MRI. As more cases are collected, we will devise methods to objectively quantify the mismatch that may arise between models derived from the two modalities. The establishment of such a framework would not only provide previously unavailable measures of shape and function, but in turn leverage the significantly wider clinical reach of echocardiography.

Download Full-text

Automated Classification of Circulating Tumor Cells and the Impact of Interobsever Variability on Classifier Training and Performance

Journal of Immunology Research ◽

10.1155/2015/573165 ◽

2015 ◽

Vol 2015 ◽

pp. 1-9 ◽

Cited By ~ 17

Author(s):

Carl-Magnus Svensson ◽

Ron Hübler ◽

Marc Thilo Figge

Keyword(s):

Random Forest ◽

Tumor Cells ◽

Circulating Tumor Cells ◽

Interobserver Variability ◽

Ground Truth ◽

Training Data ◽

Machine Learning Techniques ◽

Support Vector ◽

Data Set ◽

The Impact

Application of personalized medicine requires integration of different data to determine each patient’s unique clinical constitution. The automated analysis of medical data is a growing field where different machine learning techniques are used to minimize the time-consuming task of manual analysis. The evaluation, and often training, of automated classifiers requires manually labelled data as ground truth. In many cases such labelling is not perfect, either because of the data being ambiguous even for a trained expert or because of mistakes. Here we investigated the interobserver variability of image data comprising fluorescently stained circulating tumor cells and its effect on the performance of two automated classifiers, a random forest and a support vector machine. We found that uncertainty in annotation between observers limited the performance of the automated classifiers, especially when it was included in the test set on which classifier performance was measured. The random forest classifier turned out to be resilient to uncertainty in the training data while the support vector machine’s performance is highly dependent on the amount of uncertainty in the training data. We finally introduced the consensus data set as a possible solution for evaluation of automated classifiers that minimizes the penalty of interobserver variability.

Download Full-text

Fully Data-Driven Pseudohealthy Synthesis for Planning Valve-Sparing Aortic Root Reconstruction using Conditional Variational Autoencoders

Current Directions in Biomedical Engineering ◽

10.1515/cdbme-2020-3072 ◽

2020 ◽

Vol 6 (3) ◽

pp. 284-287

Author(s):

Jannis Hagenah ◽

Mohamad Mehdi ◽

Floris Ernst

Keyword(s):

Aortic Root ◽

Similarity Index ◽

Ground Truth ◽

Representation Learning ◽

Patient Specific ◽

Ultrasound Images ◽

Specific Geometry ◽

The Individual ◽

Native Root ◽

Original Information

AbstractAortic root aneurysm is treated by replacing the dilated root by a grafted prosthesis which mimics the native root morphology of the individual patient. The challenge in predicting the optimal prosthesis size rises from the highly patient-specific geometry as well as the absence of the original information on the healthy root. Therefore, the estimation is only possible based on the available pathological data. In this paper, we show that representation learning with Conditional Variational Autoencoders is capable of turning the distorted geometry of the aortic root into smoother shapes while the information on the individual anatomy is preserved. We evaluated this method using ultrasound images of the porcine aortic root alongside their labels. The observed results show highly realistic resemblance in shape and size to the ground truth images. Furthermore, the similarity index has noticeably improved compared to the pathological images. This provides a promising technique in planning individual aortic root replacement.

Download Full-text

Quantifying the structure of strong gravitational lens potentials with uncertainty-aware deep neural networks

Monthly Notices of the Royal Astronomical Society ◽

10.1093/mnras/staa3201 ◽

2020 ◽

Vol 499 (4) ◽

pp. 5641-5652

Author(s):

Georgios Vernardos ◽

Grigorios Tsagkatakis ◽

Yannis Pantazis

Keyword(s):

Confidence Intervals ◽

Galaxy Evolution ◽

Gravitational Lensing ◽

Probability Distributions ◽

Mass Density ◽

Ground Truth ◽

Gaussian Random Fields ◽

Training Data ◽

Gravitational Lens ◽

Data Set

ABSTRACT Gravitational lensing is a powerful tool for constraining substructure in the mass distribution of galaxies, be it from the presence of dark matter sub-haloes or due to physical mechanisms affecting the baryons throughout galaxy evolution. Such substructure is hard to model and is either ignored by traditional, smooth modelling, approaches, or treated as well-localized massive perturbers. In this work, we propose a deep learning approach to quantify the statistical properties of such perturbations directly from images, where only the extended lensed source features within a mask are considered, without the need of any lens modelling. Our training data consist of mock lensed images assuming perturbing Gaussian Random Fields permeating the smooth overall lens potential, and, for the first time, using images of real galaxies as the lensed source. We employ a novel deep neural network that can handle arbitrary uncertainty intervals associated with the training data set labels as input, provides probability distributions as output, and adopts a composite loss function. The method succeeds not only in accurately estimating the actual parameter values, but also reduces the predicted confidence intervals by 10 per cent in an unsupervised manner, i.e. without having access to the actual ground truth values. Our results are invariant to the inherent degeneracy between mass perturbations in the lens and complex brightness profiles for the source. Hence, we can quantitatively and robustly quantify the smoothness of the mass density of thousands of lenses, including confidence intervals, and provide a consistent ranking for follow-up science.

Download Full-text

ChemProps: A RESTful API enabled database for composite polymer name standardization

Journal of Cheminformatics ◽

10.1186/s13321-021-00502-6 ◽

2021 ◽

Vol 13 (1) ◽

Author(s):

Bingyin Hu ◽

Anqi Lin ◽

L. Catherine Brinson

Keyword(s):

Data Exchange ◽

Ground Truth ◽

Training Data ◽

Test Accuracy ◽

Polymer Science ◽

Data Systems ◽

Materials Informatics ◽

Related Data ◽

Weight Factors ◽

Data Points

AbstractThe inconsistency of polymer indexing caused by the lack of uniformity in expression of polymer names is a major challenge for widespread use of polymer related data resources and limits broad application of materials informatics for innovation in broad classes of polymer science and polymeric based materials. The current solution of using a variety of different chemical identifiers has proven insufficient to address the challenge and is not intuitive for researchers. This work proposes a multi-algorithm-based mapping methodology entitled ChemProps that is optimized to solve the polymer indexing issue with easy-to-update design both in depth and in width. RESTful API is enabled for lightweight data exchange and easy integration across data systems. A weight factor is assigned to each algorithm to generate scores for candidate chemical names and optimized to maximize the minimum value of the score difference between the ground truth chemical name and the other candidate chemical names. Ten-fold validation is utilized on the 160 training data points to prevent overfitting issues. The obtained set of weight factors achieves a 100% test accuracy on the 54 test data points. The weight factors will evolve as ChemProps grows. With ChemProps, other polymer databases can remove duplicate entries and enable a more accurate “search by SMILES” function by using ChemProps as a common name-to-SMILES translator through API calls. ChemProps is also an excellent tool for auto-populating polymer properties thanks to its easy-to-update design.

Download Full-text

Computational Fugue Analysis

Computer Music Journal ◽

10.1162/comj_a_00300 ◽

2015 ◽

Vol 39 (2) ◽

pp. 77-96 ◽

Cited By ~ 4

Author(s):

Mathieu Giraud ◽

Richard Groult ◽

Emmanuel Leguy ◽

Florence Levé

Keyword(s):

Complete System ◽

Ground Truth ◽

Music Analysis ◽

Automated Analysis ◽

Western Music ◽

Dynamic Visualization ◽

Ground Truth Data ◽

Dmitri Shostakovich

One of the pinnacles of form in classical Western music, the fugue is often used in the teaching of music analysis and composition. Fugues alternate between instances of a subject and other patterns and modulatory sections, called episodes. Musicological analyses are generally built on these patterns and sections. We have developed several algorithms to perform an automated analysis of a fugue, starting from a score in which all the voices are separated. By focusing on the diatonic similarities between pitch intervals, we detect subjects and countersubjects, as well as partial harmonic sequences inside the episodes. We also implemented tools to detect subject scale degrees, cadences, and pedals, as well as a method for segmenting the fugue into exposition and episodic parts. Our algorithms were tested on a corpus of 36 fugues by J. S. Bach and Dmitri Shostakovich. We provide formalized ground-truth data on this corpus as well as a dynamic visualization of the ground truth and of our computed results. The complete system showed acceptable or good results for about one half of the fugues tested, enabling us to depict their design.

Download Full-text

USING SEMANTICALLY PAIRED IMAGES TO IMPROVE DOMAIN ADAPTATION FOR THE SEMANTIC SEGMENTATION OF AERIAL IMAGES

ISPRS Annals of Photogrammetry Remote Sensing and Spatial Information Sciences ◽

10.5194/isprs-annals-v-2-2020-483-2020 ◽

2020 ◽

Vol V-2-2020 ◽

pp. 483-492

Author(s):

D. Gritzner ◽

J. Ostermann

Keyword(s):

Time Window ◽

Domain Adaptation ◽

Geographical Area ◽

Model Performance ◽

Ground Truth ◽

Semantic Segmentation ◽

Training Data ◽

Aerial Images ◽

Target Domain ◽

Training Examples

Abstract. Modern machine learning, especially deep learning, which is used in a variety of applications, requires a lot of labelled data for model training. Having an insufficient amount of training examples leads to models which do not generalize well to new input instances. This is a particular significant problem for tasks involving aerial images: often training data is only available for a limited geographical area and a narrow time window, thus leading to models which perform poorly in different regions, at different times of day, or during different seasons. Domain adaptation can mitigate this issue by using labelled source domain training examples and unlabeled target domain images to train a model which performs well on both domains. Modern adversarial domain adaptation approaches use unpaired data. We propose using pairs of semantically similar images, i.e., whose segmentations are accurate predictions of each other, for improved model performance. In this paper we show that, as an upper limit based on ground truth, using semantically paired aerial images during training almost always increases model performance with an average improvement of 4.2% accuracy and .036 mean intersection-over-union (mIoU). Using a practical estimate of semantic similarity, we still achieve improvements in more than half of all cases, with average improvements of 2.5% accuracy and .017 mIoU in those cases.

Download Full-text

On the objectivity, reliability, and validity of deep learning enabled bioimage analyses

eLife ◽

10.7554/elife.59780 ◽

2020 ◽

Vol 9 ◽

Cited By ~ 1

Author(s):

Dennis Segebarth ◽

Matthias Griebel ◽

Nikolai Stein ◽

Cora R von Collenberg ◽

Corinna Martin ◽

...

Keyword(s):

Deep Learning ◽

Signal To Noise Ratio ◽

Biological Effects ◽

Reliability And Validity ◽

Ground Truth ◽

Training Data ◽

Model Organisms ◽

Data Annotation ◽

Bioimage Analysis ◽

Model Training

Bioimage analysis of fluorescent labels is widely used in the life sciences. Recent advances in deep learning (DL) allow automating time-consuming manual image analysis processes based on annotated training data. However, manual annotation of fluorescent features with a low signal-to-noise ratio is somewhat subjective. Training DL models on subjective annotations may be instable or yield biased models. In turn, these models may be unable to reliably detect biological effects. An analysis pipeline integrating data annotation, ground truth estimation, and model training can mitigate this risk. To evaluate this integrated process, we compared different DL-based analysis approaches. With data from two model organisms (mice, zebrafish) and five laboratories, we show that ground truth estimation from multiple human annotators helps to establish objectivity in fluorescent feature annotations. Furthermore, ensembles of multiple models trained on the estimated ground truth establish reliability and validity. Our research provides guidelines for reproducible DL-based bioimage analyses.

Download Full-text