Skip to main content
2025 SCMR Conference Main banner
Final Program


Schedule at a Glance
Back

Rapid Fire Abstracts

Rapid Fire 04- Machine Learning and Artificial Intelligence II

dT1 maps: A novel approach for visualizing myocardial stress without gadolinium-based contrast agents (RF_TH_271)

Thursday, January 30, 2025
5:30 PM – 5:40 PM East Coast USA Time
Location: Blue Room PreFunction
Background: Cardiovascular magnetic resonance (CMR) stress perfusion imaging is widely used to assess ischemic heart disease, typically requiring gadolinium-based contrast agents (GBCAs). However, GBCAs pose potential risks, especially in patients with renal impairment, and add complexity to the imaging process [1]. Recent advances in stress T1 mapping offer a promising alternative, detecting changes in myocardial blood volume without the need for GBCAs [2]. Despite this, current methods are limited by the need for manual measurements and the lack of co-registration and pixel-wise visualisation of rest and stress T1 maps. This study proposes a novel transformer-based approach to automatically generate co-registered rest/stress T1 maps, facilitating pixel-wise visualization and precise quantification of stress T1 reactivity: dT1 maps.

Methods: The study retrospectively included 4,840 T1 maps from 561 subjects across two centres, using 1.5 and 3 T Siemens MRI scanners. The cohort included 103 healthy volunteers and 458 patients with various cardiovascular conditions, such as coronary artery disease, hypertrophic cardiomyopathy, and atrial fibrillation. Pharmacological stress was induced using adenosine, regadenoson, dipyridamole, and dobutamine. The proposed method leverages the Swin-Transformer architecture [3] within the VoxelMorph framework [4] to perform non-rigid registration of stress and rest T1 maps (Figure 1). This enabled the derivation of pixel-wise percentage changes in T1 values, providing a detailed map of myocardial stress reactivity. The model’s registration performance was validated by comparing the automatically generated dT1 values with manually derived values, employing intra-class correlation (ICC), Bland-Altman analysis, and mixed-model ANOVA.

Results: The transformer-based model demonstrated excellent accuracy in replicating manual stress T1 reactivity measurements, with an ICC of 0.97 and a minimal mean difference of -0.06 ± 0.74% (Figure 2). The processing time was efficient, requiring less than 1.5 seconds per sample. The model’s performance was consistent across various conditions, showing no significant dependence on pharmacological stress agents, magnetic field strengths, patient demographics, or slice positions (p ≫ 0.1).

Conclusion: The transformer-based model reliably generates automated pixel-wise stress T1 reactivity maps with high accuracy, demonstrating consistent performance across diverse conditions and patient demographics. This research introduces a novel tool to CMR, enabling detailed visualization and quantification of myocardial stress responses without the need for GBCAs, thereby advancing the non-invasive assessment of ischemic heart disease. Future work will focus on validating the dT1 values against invasive measurements across different pathologies to further establish its clinical utility.