scholarly journals Generalizability and Quality Control of Deep Learning-Based 2D Echocardiography Segmentation Models in a Large Clinical Dataset

2021 ◽  
Author(s):  
XIAOYAN Zhang ◽  
Alvaro E. Ulloa Cerna ◽  
Joshua V. Stough ◽  
Yida Chen ◽  
Brendan J. Carry ◽  
...  
Keyword(s):  
Machine Learning ◽  
Quality Control ◽  
Left Atrial Volume ◽  
Research Area ◽  
Left Ventricular ◽  
Estimation Errors ◽  
Training Scheme ◽  
Clinical Dataset ◽  
2D Echocardiography ◽  
Segmentation Models

Abstract Use of machine learning for automated annotation of heart structures from echocardiographic videos is an active research area, but understanding of comparative, generalizable performance among models is lacking. This study aimed to 1) assess the generalizability of five state-of-the-art machine learning-based echocardiography segmentation models within a large clinical dataset, and 2) test the hypothesis that a quality control (QC) method based on segmentation uncertainty can further improve segmentation results. Five models were applied to 47,431 echocardiography studies that were independent from any training samples. Chamber volume and mass from model segmentations were compared to clinically-reported values. The median absolute errors (MAE) in left ventricular (LV) volumes and ejection fraction exhibited by all five models were comparable to reported inter-observer errors (IOE). MAE for left atrial volume and LV mass were similarly favorable to respective IOE for models trained for those tasks. A single model consistently exhibited the lowest MAE in all five clinically-reported measures. We leveraged the 10-fold cross-validation training scheme of this best-performing model to quantify segmentation uncertainty for potential application as QC. We observed that filtering segmentations with high uncertainty improved segmentation results, leading to decreased volume/mass estimation errors. The addition of contour-convexity filters further improved QC efficiency. In conclusion, five previously published echocardiography segmentation models generalized to a large, independent clinical dataset—segmenting one or multiple cardiac structures with overall accuracy comparable to manual analyses—with variable performance. Convexity-reinforced uncertainty QC efficiently improved segmentation performance and may further facilitate the translation of such models.

scholarly journals Association of pericardial adipose tissue with left ventricular structure and function: a region‐specific effect?

2021 ◽  
Vol 20 (1) ◽  
Author(s):  
Jin-Seok Kim ◽  
Seon Won Kim ◽  
Jong Seok Lee ◽  
Seung Ku Lee ◽  
Robert Abbott ◽  
...  
Keyword(s):  
Adipose Tissue ◽  
Specific Effect ◽  
Left Atrial Volume ◽  
Left Ventricular ◽  
Structure And Function ◽  
Doppler Imaging ◽  
Volume Index ◽  
2D Echocardiography ◽  
Pericardial Adipose Tissue ◽  
And Function

Abstract Background The independent role of pericardial adipose tissue (PAT) as an ectopic fat associated with cardiovascular disease (CVD) remains controversial. This study aimed to determine whether PAT is associated with left ventricular (LV) structure and function independent of other markers of general obesity. Methods We studied 2471 participants (50.9 % women) without known CVD from the Korean Genome Epidemiology Study, who underwent 2D-echocardiography with tissue Doppler imaging (TDI) and computed tomography measurement for PAT. Results Study participants with more PAT were more likely to be men and had higher cardiometabolic indices, including blood pressure, glucose, and cholesterol levels (all P < 0.001). Greater pericardial fat levels across quartiles of PAT were associated with increased LV mass index and left atrial volume index (all P < 0.001) and decreased systolic (P = 0.015) and early diastolic (P < 0.001) TDI velocities, except for LV ejection fraction. These associations remained after a multivariable-adjusted model for traditional CV risk factors and persisted even after additional adjustment for general adiposity measures, such as waist circumference and body mass index. PAT was also the only obesity index independently associated with systolic TDI velocity (P < 0.001). Conclusions PAT was associated with subclinical LV structural and functional deterioration, and these associations were independent of and stronger than with general and abdominal obesity measures.

P2443Diagnostic performance of 2d-echocardiography versus cardiac catheterization in assessing left ventricular end-diastolic pressure: a meta-analysis

2019 ◽  
Vol 40 (Supplement_1) ◽  
Author(s):  
K Acosta ◽  
J Cano ◽  
E Esposo ◽  
L Shiu ◽  
A Masbang
Keyword(s):  
Cardiac Catheterization ◽  
Diagnostic Performance ◽  
Diastolic Pressure ◽  
Meta Analysis ◽  
Left Atrial Volume ◽  
Public Health Problem ◽  
Left Ventricular ◽  
Volume Index ◽  
Major Public Health Problem ◽  
2D Echocardiography

Abstract Introduction Heart failure (HF) is a major public health problem that affects 23 million people worldwide. The global incidence and prevalence rates of HF are approaching epidemiologic proportions, as evidenced by the relentless increase in the number of HF hospitalizations, the growing number of HF deaths, and the spiraling costs associated with the care of HF patients. Its diagnosis may be challenging because symptoms are nonspecific. Determination of left ventricular filling pressure (LVFP) is important to interpret equivocal symptoms so as to optimize therapy. Catheterization remains to be the gold standard; however, it is not practical to submit many patients with clinical suspicion of HF to invasive studies. Although echocardiographic indices are recommended by recent guidelines (Level IIIB), studies have shown conflicting results on its diagnostic performance. Purpose This study aims to identify the diagnostic performance of 2D-echocardiography compared with cardiac catheterization in assessing LV end-diastolic pressure (LVEDP) among adult patients with suspected HF using a meta-analysis of observational studies. Methods Eight studies with a total of 1,153 patients with suspected HF who underwent simultaneous evaluation of echocardiographic estimates of LVFP and invasive measurement of LVEDP by cardiac catheterization were included in the final analysis after extensive searching. Review Manager 5.3 was used to the assess the sensitivity and specificity of E/e' lateral, septal and average, and left atrial volume index (LAVI). Meta-Disc was applied to obtain pooled estimates, receiver operating characteristic curve (ROC) and area under curve (AUC) using a 95% confidence interval. Results Overall, pooled estimates for E/e' septal >15, E/e' lateral >12, E/e' average >13 and LAVI >34 have significant diagnostic values with pooled sensitivity of 62% (95% CI 0.54 to 0.69), 39% (95% CI 0.33 to 0.45), 81% (95% CI 0.73 to 0.87) and 53% (95% CI 0.46 to 0.61) respectively; pooled specificity of 59% (95% CI 0.53 to 0.65), 87% (95% CI 0.81 to 0.91), 72% (95% CI 0.65 to 0.78) and 69% (95% CI 0.57 to 0.79) respectively; and pooled AUC of 0.624, 0.8486, 0.8190 and 0.69 respectively. Conclusion 2D echocardiography have significant diagnostic performance compared with cardiac catheterization in assessing LVEDP among adults with suspected HF. Guidelines may be updated using this meta-analysis.

scholarly journals 0550 Insomnia is Associated with Greater Arterial Stiffness and Cardiac Dysfunction

SLEEP ◽  
2020 ◽  
Vol 43 (Supplement_1) ◽  
pp. A210-A211
Author(s):  
M E Petrov ◽  
S D Youngstedt ◽  
F Mookadam ◽  
N Jiao ◽  
L M Lim ◽  
...  
Keyword(s):  
Arterial Stiffness ◽  
Global Longitudinal Strain ◽  
Left Atrial Volume ◽  
Severity Index ◽  
Aortic Pressure ◽  
Left Ventricular ◽  
Volume Index ◽  
Left Atrial Volume Index ◽  
Normal Ranges ◽  
2D Echocardiography

Abstract Introduction Insomnia is a novel and modifiable risk factor for incident cardiovascular disease (CVD). However, identification of early markers of subclinical CVD in diagnosed insomnia is understudied. Our aim for this ongoing study is to contrast markers of cardiovascular structure and function between people with insomnia and good-sleeping controls. Methods Persons with insomnia (met ICSD-III criteria) and good sleeping controls (&lt;8 Insomnia Severity Index, mean 8-night SOL and WASO&lt;31min) were recruited from the community. Twenty-two adults (21-39y; 55% women) with no history of CVD, diabetes, inflammatory conditions, significant hypertension, or current sleep-disordered breathing (WatchPat200, Itamar Medical) were enrolled and underwent fasting cardiovascular testing. Testing included: Central augmented aortic pressure (AP) and carotid-femoral pulse wave velocity (cfPWV) for vascular stiffness; brachial artery flow mediated dilation (FMD) to assess endothelial function; and 2D echocardiography to assess ejection fraction (EF%), left ventricular global longitudinal strain (LVGLS), left atrial volume index (LAVI), mitral valve E/e’ ratio (E/e’), and lateral e’. ANCOVA models, adjusting for age, comparing persons with insomnia (n=6) to good sleeping controls (n=16) on each cardiovascular measure were conducted. Results AP (range:-5,10mmHg), cfPWV (range: 4.8-7.6m/s), EF% (range:55.0-72.0%), LVGLS (range:-26,-19%) LAVI (range:14.1-26.7mL/m2), E/e’ (range:3.2-7.8), and lateral e’ (range:0.09-0.22cm/sec) were all within normal ranges according to age and sex normative standards. Mean FMD was 8.8% (SD=4.3, range:4.3-19.8%). Age adjusted ANCOVA models indicated that the insomnia group had significantly worse cardiovascular function than good sleeping controls on cfPWV (M=6.8±0.3 vs. M=5.7±0.2; p=0.004), EF% (M=60.0±1.7 vs. M=65.2±1.0; p=0.017), LVGLS (M=-21.6±0.6 vs. M=-24.3±0.4; p=0.001), and lateral e’ (M=0.12±0.01 vs. M=0.18±0.01; p=0.003). No group differences were found for AP, FMD, LAVI, and E/e’. Conclusion Among relatively healthy young adults, people with insomnia had greater arterial stiffness and worse left ventricular systolic and diastolic functioning. Support American Academy of Sleep Medicine Foundation Focused Projects Award for Junior Investigators 179-FP-18

Deep Learning Based High-Resolution Remote Sensing Image classification

2017 ◽  
Vol 7 (10) ◽  
pp. 22
Author(s):  
Sumit Kaur
Keyword(s):  
Machine Learning ◽  
Remote Sensing ◽  
Deep Learning ◽  
Image Classification ◽  
Language Processing ◽  
Object Perception ◽  
Remote Sensing Image ◽  
Research Area ◽  
Remote Sensing Image Classification ◽  
Unsupervised Algorithms

Abstract- Deep learning is an emerging research area in machine learning and pattern recognition field which has been presented with the goal of drawing Machine Learning nearer to one of its unique objectives, Artificial Intelligence. It tries to mimic the human brain, which is capable of processing and learning from the complex input data and solving different kinds of complicated tasks well. Deep learning (DL) basically based on a set of supervised and unsupervised algorithms that attempt to model higher level abstractions in data and make it self-learning for hierarchical representation for classification. In the recent years, it has attracted much attention due to its state-of-the-art performance in diverse areas like object perception, speech recognition, computer vision, collaborative filtering and natural language processing. This paper will present a survey on different deep learning techniques for remote sensing image classification. 

scholarly journals Federated Learning in a Medical Context: A Systematic Literature Review

2021 ◽  
Vol 21 (2) ◽  
pp. 1-31
Author(s):  
Bjarne Pfitzner ◽  
Nico Steckhan ◽  
Bert Arnrich
Keyword(s):  
Machine Learning ◽  
Literature Review ◽  
Systematic Literature Review ◽  
Data Privacy ◽  
Research Area ◽  
Learning Models ◽  
Related Data ◽  
Private Data ◽  
Large Databases ◽  
Machine Learning Models

Data privacy is a very important issue. Especially in fields like medicine, it is paramount to abide by the existing privacy regulations to preserve patients’ anonymity. However, data is required for research and training machine learning models that could help gain insight into complex correlations or personalised treatments that may otherwise stay undiscovered. Those models generally scale with the amount of data available, but the current situation often prohibits building large databases across sites. So it would be beneficial to be able to combine similar or related data from different sites all over the world while still preserving data privacy. Federated learning has been proposed as a solution for this, because it relies on the sharing of machine learning models, instead of the raw data itself. That means private data never leaves the site or device it was collected on. Federated learning is an emerging research area, and many domains have been identified for the application of those methods. This systematic literature review provides an extensive look at the concept of and research into federated learning and its applicability for confidential healthcare datasets.

scholarly journals Diagnosis of left ventricular hypertrophy using non-ECG-gated 15O-water PET

2021 ◽  
Author(s):  
Jens Sörensen ◽  
Jonny Nordström ◽  
Tomasz Baron ◽  
Stellan Mörner ◽  
Sven-Olof Granstam ◽  
...  
Keyword(s):  
Method Development ◽  
Intraclass Correlation ◽  
Correlation Coefficients ◽  
Roc Curves ◽  
Left Ventricular ◽  
Intraclass Correlation Coefficients ◽  
Concentric Hypertrophy ◽  
2D Echocardiography ◽  
Positron Emission ◽  
Septal Wall Thickness

Abstract Aim To develop a method for diagnosing left ventricular (LV) hypertrophy from cardiac perfusion 15O-water positron emission tomography (PET). Methods We retrospectively pooled data from 139 subjects in four research cohorts. LV remodeling patterns ranged from normal to severe eccentric and concentric hypertrophy. 15O-water PET scans (n = 197) were performed with three different PET devices. A low-end scanner (66 scans) was used for method development, and remaining scans with newer devices for a blinded evaluation. Dynamic data were converted into parametric images of perfusable tissue fraction for semi-automatic delineation of the LV wall and calculation of LV mass (LVM) and septal wall thickness (WT). LVM and WT from PET were compared to cardiac magnetic resonance (CMR, n = 47) and WT to 2D-echocardiography (2DE, n = 36). PET accuracy was tested using linear regression, Bland–Altman plots, and ROC curves. Observer reproducibility were evaluated using intraclass correlation coefficients. Results High correlations were found in the blinded analyses (r ≥ 0.87, P < 0.0001 for all). AUC for detecting increased LVM and WT (> 12 mm and > 15 mm) was ≥ 0.95 (P < 0.0001 for all). Reproducibility was excellent (ICC ≥ 0.93, P < 0.0001). Conclusion 15O-water PET might detect LV hypertrophy with high accuracy and precision.

Plant leaf recognition with shallow and deep learning: A comprehensive study

2020 ◽  
Vol 24 (6) ◽  
pp. 1311-1328
Author(s):  
Jozsef Suto
Keyword(s):  
Machine Learning ◽  
Plant Species ◽  
Research Area ◽  
Research Papers ◽  
Plant Leaf ◽  
Leaf Characteristics ◽  
The Past ◽  
The Earth ◽  
Test Sets ◽  
Comprehensive Study

Nowadays there are hundreds of thousands known plant species on the Earth and many are still unknown yet. The process of plant classification can be performed using different ways but the most popular approach is based on plant leaf characteristics. Most types of plants have unique leaf characteristics such as shape, color, and texture. Since machine learning and vision considerably developed in the past decade, automatic plant species (or leaf) recognition has become possible. Recently, the automated leaf classification is a standalone research area inside machine learning and several shallow and deep methods were proposed to recognize leaf types. From 2007 to present days several research papers have been published in this topic. In older studies the classifier was a shallow method while in current works many researchers applied deep networks for classification. During the overview of plant leaf classification literature, we found an interesting deficiency (lack of hyper-parameter search) and a key difference between studies (different test sets). This work gives an overall review about the efficiency of shallow and deep methods under different test conditions. It can be a basis to further research.

scholarly journals Central and obstructive apneas prevalence in heart failure with reduced, mid-range and preserved ejection fraction

2020 ◽  
Vol 41 (Supplement_2) ◽  
Author(s):  
C Borrelli ◽  
P Sciarrone ◽  
F Gentile ◽  
N Ghionzoli ◽  
G Mirizzi ◽  
...  
Keyword(s):  
Heart Failure ◽  
Ejection Fraction ◽  
Left Ventricular Systolic Dysfunction ◽  
Systolic Function ◽  
Systolic Dysfunction ◽  
Cardiopulmonary Exercise Testing ◽  
Left Ventricular ◽  
Apnea Hypopnea Index ◽  
Preserved Ejection Fraction ◽  
2D Echocardiography

Abstract Background Central apneas (CA) and obstructive apneas (OA) are highly prevalent in heart failure (HF) both with reduced and preserved systolic function. However, a comprehensive evaluation of apnea prevalence across HF according to ejection fraction (i.e HF with patients with reduced, mid-range and preserved ejection fraction- HFrEf, HFmrEF and HFpEF, respectively) throughout the 24 hours has never been done before. Materials and methods 700 HF patients were prospectively enrolled and then divided according to left ventricular EF (408 HFrEF, 117 HFmrEF, 175 HFpEF). All patients underwent a thorough evaluation including: 2D echocardiography; 24-h Holter-ECG monitoring; cardiopulmonary exercise testing; neuro-hormonal assessment and 24-h cardiorespiratory monitoring. Results In the whole population, prevalence of normal breathing (NB), CA and OA at daytime was 40%, 51%, and 9%, respectively, while at nighttime 15%, 55%, and 30%, respectively. When stratified according to left ventricular EF, CA prevalence decreased from HFrEF to HFmrEF and HFpEF: (daytime CA: 57% vs. 43% vs. 42%, respectively, p=0.001; nighttime CA: 66% vs. 48% vs. 34%, respectively, p&lt;0.0001), while OA prevalence increased (daytime OA: 5% vs. 8% vs. 18%, respectively, p&lt;0.0001; nighttime OA: 20 vs. 29 vs. 53%, respectively, p&lt;0.0001). When assessing moderte-severe apneas, defined with an apnea/hypopnea index &gt;15 events/hour, prevalence of CA was again higher in HFrEF than HFmrEF and HFpEF both at daytime (daytime moderate-severe CA: 28% vs. 19% and 23%, respectively, p&lt;0.05) and at nighttime (nighttime moderate-severe CA: 50% vs. 39% and 28%, respectively, p&lt;0.05). Conversely, moderate-severe OA decreased from HFrEF to HFmrEF to HFpEF both at daytime (daytime moderate-severe OA: 1% vs. 3% and 8%, respectively, p&lt;0.05) and nighttime (noghttime moderate-severe OA: 10% vs. 11% and 30%, respectively, p&lt;0.05). Conclusions Daytime and nighttime apneas, both central and obstructive in nature, are highly prevalent in HF regardless of EF. Across the whole spectrum of HF, CA prevalence increases and OA decreases as left ventricular systolic dysfunction progresses, both during daytime and nighttime. Funding Acknowledgement Type of funding source: None

Sign in / Sign up

Export Citation Format

Share Document