Introduction

T1 mapping, a magnetic resonance imaging technique for quantifying T1 relaxation times within cardiac tissue, offers invaluable insights in the realm of cardiovascular disease prediction and diagnosis. But in the present, limited attention has been directed towards assessing the predictive potential of T1 mapping for cardiac diseases, and the existing studies in this domain have been characterized by relatively modest sample sizes¹. In our work, we precisely quantified T1 value of different AHA segments and added T1 mapping to the model, thus augmenting the accuracy of cardiac disease prediction. The integration of T1 mapping data into a comprehensive multi-gene risk assessment endows the hybrid prediction model with an expanded knowledge base, leveraging both structural and genetic insights.

Methods

Dataset

In this study, ECG, Cine, and T1 mapping images through Cardiovascular Magnetic Resonance (CMR) imaging were extracted from 50,171 participants², subjecting each data type to rigorous quality control procedures, including the removal of participants who were not included in the calculation of gene principal components and chromosome abnormalities. The final analysis included 35,971 ECG participants, as well as 26,972 for Cine³ and 34,098 for T1 mapping (Fig 2).

T1 mapping analysis

The schematic representation of the data analysis pipeline is illustrated in Fig 1. We firstly involved the utilization of nnU-Net⁴ to extract multiple segments of cardiac T1 mapping images, specifically ranging from AHA7 to AHA12. The model was trained on a private cardiac image dataset on 500 healthy subjects and applied to the UKB dataset without fine-tuning. These segments were sourced from the raw T1 map series found among the various files provided by the UK Biobank, categorized under 'experimental shMOLLI sequence images' (field id: 20214)⁵. Subsequently, we quantified these segments as distinct phenotypes, utilizing statistical metrics such as mean, median, and standard deviation (sd). Quantitative data was obtained by establishing regions of interest (ROIs) within the myocardium and implementing pixel-wise mapping to compute T1 values.

Association between T1 mapping and cardiac diseases

To assess the association between T1 mapping and some common cardiac diseases, we performed logistic regression analysis, using T1 mapping in AHA segments as a risk factor for cardiac diseases as the outcome. GWAS was conducted on these cardiac phenotypes using linear mixed-effect models, adjusted by age (at imaging), age-squared, sex, age-sex interaction, age-squared-sex interaction and the top 20 genetic PCs⁶. FUMA was used to annotate variants from the GWAS, prioritize significant genes and discover enriched pathways. Polygenic Risk Score (PRS) was utilized on each of the GWAS summary statistics of the selected cardiac diseases⁷.

Cardiac diseases prediction model

Subsequently, a hybrid prediction model was constructed, revealing superior predictive performance in specific cardiac conditions with the incorporation of T1 mapping data. To assess the performance of diverse models, we conducted predictive correlation analyses. This dataset, which including 34,098 unrelated individuals of White British descent, was divided into training (70%) and testing (30%) subsets. We leveraged ridge regression as the predictive modeling approach. The effect sizes of T1 mapping were estimated using R package. (glmnet, version 4.1-8)

Results

T1 Mapping association with common cardiac diseases

There is a significant relationship between T1 mapping and some common cardiac diseases (Fig 3). We identified a unique set of genes associated with T1 mapping that was not observed in Cine or ECG data and some pathways about heart inflammation and fibrosis (Fig 4).

Variability of myocardial segments on disease prediction

Notably, there exists substantial variability in the predictive efficacy across different cardiac segments, with AHA7 and AHA8 demonstrating notably higher beta values within the model when contrasted with other anatomical regions shown in Fig 5a and 5b.

Cardiac diseases prediction

In the context of specific diseases, the integration of T1 mapping data significantly improved the predictive capabilities of PRS. The prediction performance of T1 mapping phenotypes in relation to eight distinct cardiac diseases is presented comprehensively in Fig 5c. To illustrate, the inclusion of T1 mapping data elevates the predictive correlation of hypertension's PRS to a notable 0.457 (0.433, 0.481). Similarly, the incorporation of T1 mapping into the assessment of myocardial infarction results in an enhanced predictive correlation of 0.301 (0.273, 0.328).

Acknowledgements

This study was supported in part by the National Natural Science Foundation of China (No. 62001120, 62331021), The Royal Society (IEC\NSFC\211235) and the Shanghai Sailing Program (No. 20YF1402400, 22YF1409300).The correspondence should be sent to Prof. Chengyan Wang (Email: wangcy@fudan.edu.cn)