> ## Content Index
> Fetch the complete content index at: https://www.coios.me/llms.txt
> Use this file to discover other available public pages before exploring further.

# Automated coronary calcium scoring stratified risk better than expert reading
- URL: https://www.coios.me/automated-coronary-calcium-scoring-stratified-risk-better/
- Published: 2026-10-14T13:18:24.000Z
- Updated: 2026-10-14T17:40:27.000Z
- Description: Tested in 5,969 patients, an automated coronary calcium model stratified risk of death or myocardial infarction better than experts reading the same low-dose CT scans.
- Author: Daniel Ryan
- Tags: Data, models and systems, AI in medicine, Heart and stroke, #preprint, #new, #item, #source-2026-10

A deep-learning model scored coronary artery calcium on the low-dose CT taken alongside myocardial perfusion imaging, which is rarely reported. No low-dose scan was labelled in training; expert scores from gated CT were carried over to degraded copies of those same scans. The model was trained at one centre and tested in 5,969 patients at two others, median age 64\. Agreement was higher than for models trained on gated images alone, with Cohen's kappa of 0.86 against 0.81\. Net reclassification for risk of death or myocardial infarction exceeded that of expert reading of the same scans by 0.089 (95% CI 0.051 to 0.125).

*Why it is interesting: The model was taught entirely from a different, better scan type, yet read calcium scores better than experts on the routine scans. Those low-dose scans are already acquired in most PET and many SPECT perfusion studies.*

Source

[medRxiv, 12 October 2026](https://doi.org/10.64898/2026.07.02.26356002?ref=coios.me)

Paper

Cross-Domain Knowledge Transfer from Expert-Annotated Gated CT via Synthetic Ungated CT Improves Coronary Artery Calcium Scoring on CT Attenuation Correction Scans

Authors

Shanbhag A, Miller RJ, Killekar A, Marcinkiewicz AM, Zhou J, Lemley M et al.

DOI

10.64898/2026.07.02.26356002

Type

Preprint

Design

Deep-learning model trained on synthetic ungated CT derived from labelled gated scans, externally tested in 5,969 patients at two centres

Verdict

New finding

Driver

[AI in medicine](https://www.coios.me/d-ai-medicine/)

Driver

[Heart and stroke](https://www.coios.me/d-cardiovascular/)