MétaCan
Menu
← Back to cohort

Abstract 16082: Machine Learning to Improve Left Ventricular Scar Quantification in Hypertrophic Cardiomyopathy Patients

2020· article· en· W3160047223 on OpenAlexaffabout
Zeinab Navidi Ghaziani, Jesse Sun, Raymond H. Chan, Harry Rakowski, Martin S. Maron, Ethan J. Rowin, Bo Wang, Wendy Tsang

Bibliographic record

VenueCirculation · 2020
Typearticle
Languageen
FieldMedicine
TopicCardiac Imaging and Diagnostics
Canadian institutionsToronto General HospitalUniversity Health Network
Fundersnot available
KeywordsHypertrophic cardiomyopathyArtificial intelligenceMedicineConvolutional neural networkDeep learningSegmentationMagnetic resonance imagingRobustness (evolution)Sørensen–Dice coefficientVentricleImage segmentationPattern recognition (psychology)Computer scienceMachine learningRadiologyCardiology

Abstract

fetched live from OpenAlex

Introduction: Accurate and reproducible scar quantification of late gadolinium enhancement (LGE) images from cardiac magnetic resonance imaging (CMR) is important in risk stratifying hypertrophic cardiomyopathy (HCM) patients. Previous machine learning algorithms for CMR LGE quantification deployed three-dimensional convolutional neural network (CNN) architecture, which required image cropping and custom graphic processing units (GPUs) to function, thus limiting their general applicability. We aim to develop a deep two-dimensional (2D) CNN model that contours the left ventricle (LV) endo- and epicardial borders and quantifies LGE. Hypothesis: We hypothesize that a deep 2D CNN model, which uses commercially available GPUs, could be used to efficiently and accurately contour LV endo- and epicardial borders and quantify CMR LGE in HCM patients. Methods: We retrospectively studied 296 HCM patients (2423 images) from the University Health Network (Toronto, Canada) and Tufts Medical Center (Boston, USA). LGE images were manually segmented by an expert reader. Scar was defined as 5 standard deviations higher than the mean of the annotated normal region pixels. A 2D U-net CNN variant was used to train a model on 80% of the datasets. Testing was performed on the remaining 20%. We applied a 5-folds cross validation algorithm for training to improve model robustness. Model performance was assessed using the Dice Similarity Coefficient (DSC). Results: We were able to develop a deep learning model that could successfully perform both LV segmentation and scar quantification using a generally available GPU card. Our algorithm did not require image cropping and processed one image every 60 milliseconds. DSC scores averaged across the 5-folds was excellent at 0.89+0.22 for the endocardium and 0.81+0.17 for the epicardium, and good at 0.57+0.31 for scar. Conclusions: Using novel 2D CNN methods, we have successfully developed an automatic algorithm that rapidly provides LV endo- and epicardial contours and scar quantification on LGE CMR images that is superior to previously published studies. Unlike previous algorithms, our program does not require the use of custom CPUs or image cropping, potentially allowing it to be integrated into routine clinical practice.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

How this classification was reachedexpand

Full frame machine prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.

metaresearch head score (Codex)0.001
metaresearch head score (Gemma)0.003
Version: metacan-v3-hybrid-931329e0061cValidation status: machine_predicted_unvalidated
Candidate categoriesnone
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Observational · Consensus signal: none
GenreCandidate signal: Empirical · Consensus signal: Empirical
Teacher disagreement score0.002
Threshold uncertainty score0.006

Distilled classifier scores by category (both heads)

CategoryCodexGemma
Metaresearch0.0010.003
Meta-epidemiology (narrow)0.0000.000
Meta-epidemiology (broad)0.0000.000
Bibliometrics0.0010.000
Science and technology studies0.0000.000
Scholarly communication0.0000.000
Open science0.0000.000
Research integrity0.0010.000
Insufficient payload (model declined to judge)0.0010.000

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.015
GPT teacher head0.244
Teacher spread0.229 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.

The models applied no category: nothing in the taxonomy fit this work.
Study designObservational
Domainnot available
GenreEmpirical

How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".

Quick stats

Citations0
Published2020
Admission routes2
Has abstractyes

Explore more

Same venueCirculation→Same topicCardiac Imaging and Diagnostics→French-language works237,207→