Decoding the Age Within

Chronological age, measured from the time of birth, has long been a standard metric in healthcare and aging research. However, it is an imperfect measure, lacking the nuanced information provided by biological age, which considers various genetic and environmental factors. Biological age estimates are generated through mathematical models that use biomarkers as predictors and chronological age as the output. The difference between biological and chronological age—known as the “age gap”—serves as a complementary indicator of aging, offering additional insights beyond the limitations of chronological age alone.

The utility of the "age gap" becomes evident when examining its correlation with specific exposures, like lifestyle choices or pre-existing health conditions. For instance, in brain age estimation, neuroimaging biomarkers are used as predictors. The "brain age gap" between model-predicted age and chronological age serves as an avenue to study the impact of genetic and environmental factors on brain aging. Similarly, this "age gap" concept can be extended to other organs and systems, including, as the Osaka study suggests, the chest.

New paper published in The Lancet Healthy Longevity suggest that AI can use chest radiographs to predict age with astonishing accuracy and that disparities between AI-estimated age and actual age could be indicators of chronic illness.

The research team constructed a deep learning-based AI model trained on 67,099 chest radiographs from 36,051 healthy individuals. The focus of the model was biomarker modelling, specifically examining saliency maps of AI findings on chest radiographs. By minimizing overfitting through the use of multi-institutional data, the model achieved a remarkable correlation coefficient of 0.95 between AI-estimated age and actual age.

The saliency maps were particularly telling. The "hot regions" on the map coincided with areas known for age-related markers such as calcification and aortic arch tortuosity. Similarly, "cold regions" aligned with younger ages and were consistent with previously reported signs of emphysematous changes in the lower lung fields. The biomarker modeling phase thus successfully validated many age-related findings in chest radiographs known from prior research.

An additional set of 34,197 chest radiographs from patients with known diseases were analyzed. The data revealed that disparities between the AI-estimated age and chronological age were significantly correlated with chronic diseases like hypertension, hyperuricemia, and chronic obstructive pulmonary disease.

CNV clocks are suggested as unique aging biomarkers, distinct from DNA methylation-based epigenetic clocks. An intriguing alternative aging model, leveraging the dynamics of Copy Number Variations (CNVs) and their corresponding gene expressions, is inspired by the body's post-partum recuperative processes. Although both pregnancy and aging exert distinct types of biological stress on the body, their overlapping cellular mechanisms could offer valuable insights into reversing age-related changes. 

One of the emerging tools in biological age estimation is the use of epigenetic clocks that can track both chronological and biological age. These clocks rely on advanced methodologies like EpiScores and non-linear DNAm effects, leveraging large datasets for improved prediction accuracy. The importance of these tools is not just in their predictive power, but also in the insights they provide into complex traits and adverse health outcomes.

However, the field faces challenges, particularly the limited generalizability across studies due to variations in samples, predictors, and modelling methods. These variations underline the need for exact reporting in scientific studies to ensure methodological transparency and allow for cross-study comparisons.


Yasuhito Mitsuyama, Toshimasa Matsumoto, Hiroyuki Tatekawa, Shannon L Walston, Tatsuo Kimura, Akira Yamamoto, Toshio Watanabe, Yukio Miki, Daiju Ueda. Chest radiography as a biomarker of ageing: artificial intelligence-based, multi-institutional model development and validation in Japan. The Lancet Healthy Longevity, August 17, 2023 DOI: 10.1016/S2666-7568(23)00133-2

Salih A, Nichols T, Szabo L, Petersen SE, Raisi-Estabragh Z. Conceptual Overview of Biological Age Estimation. Aging Dis. 2023 Jun 1;14(3):583-588. doi: 10.14336/AD.2022.1107. PMID: 37191413; PMCID: PMC10187689.

Babyn PS, Adams SJ. AI analysis of chest radiographs as a biomarker of biological age. The Lancet Healthy Longevity. 2023 Aug 16.


Popular posts from this blog

AI in Elderly Healthcare: The Promise and The Bias

Neuropathological Features of Age-Related Brain Diseases

HippoCamera for Aging Brain