The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2634
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-12T12:02:08.774156Z
status: published
lens: trace
sector: health
headline: Machine learning-driven ECG screening for early coronary artery disease detection
dek: BackgroundElectrocardiography (ECG) is inexpensive and widely available, yet its ability to detect early and nonobstructive coronary artery disease (CAD) is limited when interpreted using conventional criteria. We evaluated whether a deep learning model applied to standard 12-lead-ECG images can identify angiographically confirmed CAD and enable clinically meaningful risk stratification in a real-world population.MethodsIn this multicentric study, 15,890 ECG-angiography pairs (12-lead ECG paired with coronary an…
gain_title: In a multicentric study, a deep learning model reading routine ECG images detected 618 of 708 angiography-confirmed coronary artery disease cases (87% sensitivity) on a 1,339-sample test set, with a ROC-AUC of 0.80.
problem_title: In the same multicentric ECG study, the deep learning model correctly classified only 317 of 631 people without coronary artery disease (50% specificity) on the test set, and its sensitivity fell to 0.67 on the later validation cohort of 2,168 samples.
trace_subject: Deep learning ECG image model for detecting coronary artery disease
gain_reading: In a multicentric study, a deep learning model reading routine ECG images detected 618 of 708 angiography-confirmed coronary artery disease cases (87% sensitivity) on a 1,339-sample test set, with a ROC-AUC of 0.80.
gain_evidence: sensitivity was 87% (618/708 CAD cases detected) | ROC-AUC was 0.80 and PR-AUC 0.83
problem_reading: In the same multicentric ECG study, the deep learning model correctly classified only 317 of 631 people without coronary artery disease (50% specificity) on the test set, and its sensitivity fell to 0.67 on the later validation cohort of 2,168 samples.
problem_evidence: specificity 50% (317/631 non-CAD correctly classified) | sensitivity 0.67, specificity 0.78, PPV 0.84, and NPV 0.57
quick_read: A multicentric study trained a DenseNet-121 neural network on 15,890 paired 12-lead ECG images and coronary angiography or coronary CT angiography results to predict coronary artery disease (CAD). On the test set of 1,339 samples, the model reached a ROC-AUC of 0.80 and caught 618 of 708 CAD cases.

The trade-off was many false alarms: at the 0.50 threshold, specificity was 50%, meaning 317 of 631 people without CAD were correctly classified. On a later, independent cohort of 2,168 samples collected prospectively from February to May 2024, ROC-AUC was 0.81, but sensitivity fell to 0.67 while specificity rose to 0.78.

The authors conclude that routinely acquired ECG images may support low-cost, scalable CAD risk stratification. The negative predictive value of 0.57 on the temporal cohort suggests a negative result would not reliably rule out disease in a population where 64% had CAD.
limitation: Specificity was only 50% on the test set, and on the temporal validation cohort sensitivity dropped to 0.67 with an NPV of 0.57, so the model's operating balance varied between datasets.
tag: Dual reading
key_points: The study included 15,890 ECG-angiography pairs collected between November 2021 and January 2024 across multiple centres. | The model was trained on 13,299 samples, validated on 1,002 and tested on 1,339, with CAD prevalence of 52.9% in the test set. | Test-set ROC-AUC was 0.80 and PR-AUC 0.83; sensitivity was 87% and specificity 50% at the 0.50 threshold. | A prospectively collected temporal validation cohort of 2,168 samples (64% CAD) gave ROC-AUC 0.81 and PR-AUC 0.83. | On that cohort, sensitivity was 0.67, specificity 0.78, PPV 0.84 and NPV 0.57. | Simulations of varying disease prevalence were used to check how PPV and NPV trends held across different populations.
rundown: Standard ECG is cheap and widely available but performs poorly at detecting early and nonobstructive coronary artery disease when read using conventional criteria. The researchers asked whether a convolutional neural network applied to ECG images alone could identify angiographically confirmed disease in a real-world population.

The model, a DenseNet-121, was trained on ECG images paired with coronary angiography or coronary CT angiography results. Performance was assessed with ROC-AUC, PR-AUC and standard diagnostic metrics, and prevalence-dependent behaviour was summarised with PPV and NPV prevalence curves (AUCppv and AUCnpv).

Results were consistent in overall discrimination (ROC-AUC 0.80 on the test set, 0.81 on the later cohort) but the balance between catching disease and avoiding false positives shifted between datasets. The authors describe the findings as supporting risk stratification rather than a standalone diagnosis.
sources:
- peer_reviewed | Asian Cardiovascular and Thoracic Annals | https://doi.org/10.1177/02184923261503901 | 2026-11-11
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

e873fd8bb698783f203a0b99265ca72476ef0e65891c309f54dfdc332a44c535

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000