Age-related diseases like fatty liver disease, cirrhosis, and type 2 diabetes become more prevalent with age. Predicting abdominal age and identifying risk factors for accelerated abdominal aging could lead to interventions that delay the onset of these diseases. While biological age predictors exist for various organs (brain, heart, etc.), none have used abdominal MRIs (liver and pancreas). This research addresses this gap by building the first abdominal age predictor, leveraging the extensive data available in the UK Biobank. The study aims to determine if abdominal age can be accurately predicted from liver and pancreas MRIs using deep learning, to identify the anatomical features driving the predictions, to quantify the heritability of accelerated abdominal aging, and to explore the association between accelerated abdominal aging and a wide range of genetic and non-genetic factors. The significance lies in potentially identifying early indicators of age-related abdominal diseases and developing targeted preventative strategies.

The literature highlights age-related changes in abdominal organs like the liver (cellular and macroscopic changes, increased vulnerability to diseases) and pancreas (fibrosis, atrophy, increased susceptibility to diabetes and cancer). Existing biological age predictors utilize various data types including brain and heart MRIs, electrocardiograms, ultrasound images, X-rays, blood samples, DNA methylation, and more. However, no prior research has explored the prediction of age using liver and pancreas MRIs. This study builds upon this existing work by introducing a novel approach to abdominal aging assessment.

The study used 45,552 liver MRIs and 36,784 pancreas MRIs from UK Biobank participants (aged 37-82). Deep convolutional neural networks (InceptionResNetV2 and InceptionV3) with transfer learning were employed to predict age. The images underwent preprocessing (adaptive histogram equalization, cropping). Data augmentation techniques were used to prevent overfitting. Model performance was evaluated using R-squared, mean absolute error (MAE), and root mean squared error (RMSE). Attention maps were generated to identify image features driving age prediction. Genome-wide association studies (GWAS) were conducted to assess the heritability of accelerated abdominal aging and identify associated single nucleotide polymorphisms (SNPs). X-wide association studies (XWAS) were performed to examine associations with biomarkers, clinical phenotypes, diseases, environmental, and socioeconomic variables. Ensemble models were created to combine predictions from different algorithms and image preprocessing methods. Statistical biases in age prediction residuals were corrected using linear regression. Genetic correlations between abdominal aging and other organ systems' aging were also computed.

The ensemble model accurately predicted abdominal age (AbdAge) with an R-squared of 73.3 ± 0.6 and MAE of 2.94 ± 0.03 years. Attention maps revealed that predictions were driven by liver and pancreas features, along with surrounding organs and tissues. GWAS analysis estimated the heritability of accelerated abdominal aging at 26.3 ± 1.9% and identified 16 genetic loci associated with this phenotype, including genes implicated in age-related macular degeneration (PLEKHA1, ARMS2, HTRA1) and EFEMP1. XWAS identified numerous associations with non-genetic factors. Body impedance, blood pressure, and pulse wave analysis biomarkers were strongly associated with accelerated abdominal aging, while hand grip strength, cognitive tests, and bone density were associated with decelerated aging. Clinical phenotypes like chest pain, breathing difficulties, and poor general health were linked to accelerated aging. Cardiovascular and pulmonary diseases were also associated. Smoking, sedentary behavior, and specific alcohol types were associated with accelerated aging, while higher levels of physical activity, certain dietary factors, and higher socioeconomic status were associated with decelerated aging. Liver and pancreas accelerated aging were significantly phenotypically, genetically, and environmentally correlated. AbdAge showed moderate correlation with cardiac aging but was largely independent of other organ system aging predictors.

The high accuracy of the AbdAge predictor demonstrates the potential of using deep learning and abdominal MRIs to assess abdominal aging. The attention maps suggest the model captures general abdominal aging rather than organ-specific aging. The findings highlight the multifactorial nature of abdominal aging, involving genetic predisposition, biomarkers reflecting metabolic health, clinical presentations of various organ systems, environmental exposures, and socioeconomic factors. The strong correlations between liver and pancreas aging suggest shared underlying mechanisms. The associations with cardiovascular, pulmonary, and neurological health indicate that accelerated abdominal aging may be a marker of overall health decline. Future research should explore the specific mechanisms linking these factors and investigate whether interventions targeting these factors could modify abdominal aging trajectory.

This study introduces AbdAge, a robust deep-learning based predictor of abdominal age from liver and pancreas MRIs. The findings underscore the complexity of abdominal aging and its relationship to multiple biological systems. The identified genetic loci and non-genetic factors could serve as targets for future interventions, and AbdAge offers a valuable tool for assessing the impact of rejuvenating therapies on abdominal health. Future work could focus on improved image segmentation to create organ-specific age predictors, investigate the role of stromal stellate cells in abdominal aging, explore the causal relationships between AbdAge and various diseases using Mendelian randomization, and further characterize the multidimensionality of aging.

The study's reliance on the UK Biobank dataset may limit the generalizability of findings to other populations. The cross-sectional nature of the data prevents causal inference. The attention maps provide insights but do not definitively identify the precise biological mechanisms. The focus on liver and pancreas MRIs may not capture the full spectrum of abdominal aging changes.