PulseTrendingJournal ClubResearchersJournalsExplore
Instagram
HomeTrendingJournal ClubExplore
Synapse
⌘+K
Synapse
January 14, 2026European Heart Journal - Digital HealthOpen Access

Explainable visual transformer based scoring of CAD-RADS from coronary CT angiography multiplanar projections

View Full Paper
Ask AI
Bookmark
Share

Why the study?

Interpreting CCTA scans and assigning CAD-RADS scores is resource-intensive and operator-dependent, driving demand for automated and transparent strategies to support routine CAD screening.

Does an automated Multi-Axis Vision Transformer model accurately predict CAD-RADS scores and the need for follow-up investigation from CCTA?

Population

253 patients with standard CCTA

Comparison

Fine-tuned MaxViT AI model vs conventional CNN and attention-based baselines

Design

AI model development and validation study

Key result

The Multi-Axis Vision Transformer model achieved an AUC of 0.93 and 0.88 accuracy in CAD-RADS scoring from coronary CT angiography in a cohort of 253 patients.

Authors

EPE ParimbelliMCMattia ChiesaTBT M Buonocore

Discussion

Loading...

Member takes

Overview

May facilitate automated CAD-RADS scoring; leaves open prospective multicenter validation before clinical use.

Key Points

  • Evaluate an automated scoring method using visual transformers for CAD-RADS from coronary CT angiography.
  • Developed a web-based platform for CAD-RADS scoring without additional annotations.
  • Utilized 2D images of coronary arteries to create tri-channel composite images for analysis.
  • Fine-tuned a Multi-Axis Vision Transformer for categorizing CAD risk and identifying further investigations.
  • Achieved an AUC of 0.93 for three-class CAD risk classification and 0.87 for follow-up investigation decision.
  • Classification accuracy reached 0.88 with good usability for non-technical users.
  • Demonstrated superior performance compared to conventional CNN and attention-based models.

Structured PICO

Does an automated Multi-Axis Vision Transformer model accurately predict CAD-RADS scores and the need for follow-up investigation from CCTA?

P
Population
253 patients undergoing Coronary CT Angiography (CCTA)
I
Intervention
Automated CAD-RADS scoring using a fine-tuned Multi-Axis Vision Transformer (MaxViT) model with visual and textual explainability
C
Comparator
Conventional CNN and attention-based baselines
O
Outcome
Three-class CAD risk prediction (low [CAD-RADS 0], intermediate [1-3], or high [4+]) and binary decision for follow-up investigationsurrogate

An explainable visual transformer model demonstrated high accuracy in automating CAD-RADS scoring and identifying the need for further invasive investigation from standard CCTA.

Cite This Study

Parimbelli et al. (2026) studied this question. The Multi-Axis Vision Transformer model achieved an AUC of 0.93 and 0.88 accuracy in CAD-RADS scoring from coronary CT angiography in a cohort of 253 patients.

synapsesocial.com/papers/69671985c0d1e3cfbfce8ec0https://doi.org/10.1093/ehjdh/ztaf143.044
View Full Paper
Ask AI
Bookmark
Share

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1Patient-level CAD-RADS scoring from coronary radiomic features2026 · 1 citations
  2. 2Optimization of pre-test probability models for obstructive coronary artery disease using explainable artificial intelligence2025
  3. 3Potential of artificial intelligence to differentiate between non-obstructive and obstructive coronary artery disease on coronary CT angiography2024
  4. 4Clinical Performance Evaluation of an Artificial Intelligence–Based Tool for Predicting the Presence of Obstructive Coronary Artery Disease: Protocol for a Cohort Observational Study (Preprint)2025
  5. 5The potential of artificial intelligence in the assessment of coronary artery stenosis in follow-up CT examinations2024