Machine learning applied to abdominal MRI scans identified splenic radiomic features associated with coronary artery disease, with genome-wide association analysis linking these features to 219 loci including 9p21, the strongest yet mechanistically elusive CAD locus.
Key Findings
Results
Deep learning extracted 107 splenic radiomic features from abdominal MRI scans, of which 10 were associated with coronary artery disease in the UK Biobank.
Deep learning was applied to abdominal MRI scans of 42,059 UK Biobank participants and 2,745 Mass General Brigham Biobank (MGBB) participants.
107 splenic radiomic features were extracted in total.
10 of these features from the UK Biobank cohort were associated with CAD.
The study used a framework combining deep learning with genomics to investigate splenic involvement in CAD.
Results
Genome-wide association analysis of CAD-associated splenic features identified 219 genomic loci.
The genome-wide association analysis was conducted on the 10 CAD-associated splenic radiomic features.
219 loci were identified in total.
Notable loci included 9p21, described as 'the strongest yet mechanistically elusive CAD locus.'
The analysis was performed in UK Biobank participants with available MRI and genomic data.
Results
Variants at the 9p21 locus were associated with splenic features reflecting heterogeneity of continuous texture regions.
9p21 is characterized in the paper as 'the strongest yet mechanistically elusive CAD locus.'
The specific splenic feature associated with 9p21 variants was 'run-length nonuniformity,' which reflects heterogeneity of continuous texture regions.
This association provides a potential mechanistic link between 9p21 variants and CAD through splenic biology.
The connection between 9p21 and the spleen had not previously been described.
Results
External clinical validation in the MGBB cohort highlighted challenges in translating research MRI findings to routine clinical practice.
The MGBB cohort included 2,745 participants with clinical abdominal MRI scans.
Research MRI findings were consistent internally within the UK Biobank.
External validation revealed challenges due to 'variability in imaging protocols and greater clinical heterogeneity among patients.'
The authors emphasize 'translational gaps between research and clinical radiomics' as a key limitation.
Background
Circulating hematopoietic cells influence CAD risk separately from traditional risk factors, and the spleen as a key regulating organ of the hematopoietic system was previously understudied in this context.
The paper notes that 'circulating hematopoietic cells influence risk for CAD separately from traditional risk factors.'
The spleen is described as 'a key regulating organ' of the hematopoietic system.
Despite advances in managing traditional risk factors, CAD 'remains the leading cause of mortality.'
The authors frame the spleen as 'optimally suited for unbiased radiologic investigations toward mechanistic insights' into CAD.
What This Means
This research suggests that the spleen — an organ not traditionally associated with heart disease — may play a role in coronary artery disease (CAD), the leading cause of death worldwide. Using artificial intelligence to analyze abdominal MRI scans from over 42,000 participants in the UK Biobank, the researchers identified patterns in the spleen's texture and structure that were linked to CAD. They then searched the genome for genetic variants associated with these splenic patterns and found 219 regions of the genome, including a well-known but poorly understood genetic region called 9p21, which is the strongest known genetic risk factor for CAD. The finding that 9p21 variants are linked to specific splenic texture features (reflecting how uniform or irregular the spleen's internal regions are) offers a potential new clue about how this genetic region contributes to heart disease risk through the spleen and the blood cell-producing system it regulates.
The study also tested whether these findings could be applied in a real-world clinical setting using MRI scans from patients at Mass General Brigham hospitals. While the research-grade MRI results were consistent within the UK Biobank, applying the same approach to routine clinical scans proved more difficult because clinical MRI scans vary in how they are taken and because patients seen in hospitals tend to be more medically complex. This highlights an important gap between discoveries made using standardized research data and the practical application of such findings in everyday medical care.
Overall, this research suggests a framework for using AI-driven image analysis combined with genetics to uncover new biological pathways in disease — in this case, pointing toward the spleen's potential involvement in heart disease. However, the authors caution that significant work remains to bridge the gap between research findings and clinical use, particularly in making radiomic analyses robust enough to work reliably across different hospitals and imaging settings.