- The new AI can precisely estimate biological age and major risk factors related to health.
- It analyzes data collected by smartphones and wearable devices.
- Developers have released an app that observes how your daily activity pattern affects your life expectancy.
Many biological factors like gene expression, DNA and blood circulation levels show close correlation with age. However, genome profile or large-scale biochemical are quite hard and expensive for any applications beyond scientific research.
To make things easy, developers at biotech company GERO, and Moscow Institute of Physics and Technology, Russia, have built an artificial intelligence system that can estimate biological age and major risk factors related to health. It works by analyzing the data collected by smartphones and wearable devices.
AI has already demonstrated phenomenal performance in pattern recognition, speech recognition, visual object identification and other areas. In fact, neural networks are being utilized in the medical field to provide personalized treatment and make drugs. Inspired by these tools, researchers have now developed a new system that can provide accurate health information based on physical activities.
Modern operating systems of handheld and wearable devices enable collection and cloud-storage of personal activity record without disturbing user’s daily routine. And this is done on a very large scale – for billions of people. The AI utilizes these records to continuously monitor health related risks and provide feedback in real-time.
How Did They Do It?
Researchers extracted 4 years (2003 to 2006) of clinical data and physical activity records from NHANES (National Health and Nutrition Examination Survey). Then they trained the neural network on 1-week records to estimate mortality risk and biological age.
They compared 3 increasingly precise biological age models –
- Multivariate linear regression
- Unsupervised Principal Components Analysis (PCA)
- Deep Convolutional Neural Network (CNN)
Researchers discovered that the supervised method or CNN unraveled most of the biological motion patterns and established their relation to lifespan and general health information. The algorithm outperformed all existing models of mortality risks and biological age running on the same data.
The team has developed an iOS app that sees (using phone’s accelerometer) how user’s daily activity patterns affect their life expectancy.
Moreover, in their previous work, the team adopted transition matrix elements, aggregated descriptors, and a simple form of quantile normalization to demonstrate that the AI trained on NHANES data could be used to estimate health risks in United Kingdom Biobank.
Some health insurance companies have already started to offer discounts based on user’s physical activities, monitored via wearable devices.
According to the developers, the algorithm can be further improved to provide more accurate risk models. Combining latest machine learning techniques with aging theory will yield even better health models to reduce longevity risks in insurance and help in retirement planning. The AI could also contribute in the development of anti-aging therapies and future clinical trials.