Sensitivity-oriented fusion of resting ECGs and structured CCTA reports for identifying angiography-defined severe right coronary artery stenosis: a single-centre retrospective study

Background: Failure to identify severe right coronary artery (RCA) stenosis may delay invasive assessment and treatment planning. Structured coronary computed tomography angiography (CCTA) reports provide anatomical screening information, but report-derived stenosis summaries do not capture electrophysiological consequences. We fused resting 12-lead electrocardiograms (ECGs) with structured CCTA report features in a multimodal model. We then evaluated sensitivity-oriented identification of RCA diameter stenosis of at least 70% in coronary angiography (CAG) reports. Results: This single-centre retrospective study included 1,904 ECG records from 760 patients who underwent ECG, CCTA, and CAG. CAG reports provided the reference labels. The model combined a one-dimensional residual network, axial multi-head self-attention, lead attention, and feature-wise linear modulation by report-derived stenosis features from four coronary arteries. Performance was estimated using internal patient-grouped fivefold cross-validation. At the patient level, the multimodal model achieved sensitivity of 0.849±0.037 and specificity of 0.690±0.037. Its area under the receiver operating characteristic curve (AUC) was 0.831±0.020, with accuracy of 0.764±0.028 and F1 score of 0.761±0.051. The structured-report model achieved an AUC of 0.794, sensitivity of 0.391, and specificity of 0.837. Thus, multimodal fusion increased sensitivity while reducing specificity. The ECG-only model achieved an AUC of 0.574. Ablation analyses associated CCTA fusion, temporal-decay weighting, and attention modules with internal performance. An ancillary post hoc analysis reclassified 38 of 66 CCTA false-positive patients as negative, but it did not establish clinical net benefit. Conclusions: In this selected single-centre cohort, ECG and structured CCTA report fusion produced a high-sensitivity operating profile for angiography-defined severe RCA stenosis, with a corresponding reduction in specificity. The model may warrant evaluation as a sensitivity-oriented secondary screening aid, but it cannot yet be used to rule out disease or claim prevention of missed diagnoses. Independent multicentre validation, prespecified threshold locking, calibration, and prospective clinical-utility assessment are required before implementation.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-20
DOI
https://doi.org/10.21203/rs.3.rs-10387984/v1
Primary Topic
ECG Monitoring and Analysis
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Sensitivity-oriented fusion of resting ECGs and structured CCTA reports for identifying angiography-defined severe right coronary artery stenosis: a single-centre retrospective study

Shenqi Jing, Tao Shan, Yue He, Yong Li et al.
1 links
Research Square
ECG Monitoring and Analysis
preprint

Sensitivity-oriented fusion of resting ECGs and structured CCTA reports for identifying angiography-defined severe right coronary artery stenosis: a single-centre retrospective study

Shenqi Jing, Tao Shan, Yue He, Yong Li, Qiuyang Yang, Yun Liu
preprint en

Abstract

Background: Failure to identify severe right coronary artery (RCA) stenosis may delay invasive assessment and treatment planning. Structured coronary computed tomography angiography (CCTA) reports provide anatomical screening information, but report-derived stenosis summaries do not capture electrophysiological consequences. We fused resting 12-lead electrocardiograms (ECGs) with structured CCTA report features in a multimodal model. We then evaluated sensitivity-oriented identification of RCA diameter stenosis of at least 70% in coronary angiography (CAG) reports. Results: This single-centre retrospective study included 1,904 ECG records from 760 patients who underwent ECG, CCTA, and CAG. CAG reports provided the reference labels. The model combined a one-dimensional residual network, axial multi-head self-attention, lead attention, and feature-wise linear modulation by report-derived stenosis features from four coronary arteries. Performance was estimated using internal patient-grouped fivefold cross-validation. At the patient level, the multimodal model achieved sensitivity of 0.849±0.037 and specificity of 0.690±0.037. Its area under the receiver operating characteristic curve (AUC) was 0.831±0.020, with accuracy of 0.764±0.028 and F1 score of 0.761±0.051. The structured-report model achieved an AUC of 0.794, sensitivity of 0.391, and specificity of 0.837. Thus, multimodal fusion increased sensitivity while reducing specificity. The ECG-only model achieved an AUC of 0.574. Ablation analyses associated CCTA fusion, temporal-decay weighting, and attention modules with internal performance. An ancillary post hoc analysis reclassified 38 of 66 CCTA false-positive patients as negative, but it did not establish clinical net benefit. Conclusions: In this selected single-centre cohort, ECG and structured CCTA report fusion produced a high-sensitivity operating profile for angiography-defined severe RCA stenosis, with a corresponding reduction in specificity. The model may warrant evaluation as a sensitivity-oriented secondary screening aid, but it cannot yet be used to rule out disease or claim prevention of missed diagnoses. Independent multicentre validation, prespecified threshold locking, calibration, and prospective clinical-utility assessment are required before implementation.

Research Square
Nanjing University of Posts and Telecommunications (CN), Jiangsu Province Hospital (CN)
Nanjing Medical University, National Key Research and Development Program of China
Good health and well-being
ECG Monitoring and Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon