Neural network that determines the age of the blood test - the development of scientists of the ITMO University
Scientists from the Computer Technologies laboratory of the ITMO University as part of an international group of researchers developed the Aging.AI system — an age determination algorithm based on the results of a basic blood test.
Unlike other developments, this one is more versatile and accurate.
In the course of the study, scientists analyzed data sets of people of different nationalities and showed that such an approach has more predictive accuracy than developments based on data of people belonging to the same population.
')
Read more about this study in our material.
Photo Aging.AI
The definition of biological age is one of the promising areas of research related to the study of aging. Information about a person’s biological age (which may not always coincide with the chronological, or, in other words, “passport age”) can be used in the future when prescribing drugs or deciding whether a person can be a donor.
British researchers working in this area say that the ability to assess a person’s biological age (and compare it with chronological) can change not only medicine, but other industries as well, such as insurance, where age is an important indicator of scoring in some cases.
A simpler, but also non-trivial task is to search for biomarkers that allow one to estimate the chronological age. Such biomarkers, for example, can be a useful tool for assessing the therapeutic efficacy of a drug or procedure aimed at “slowing down aging”. It was this task - the determination of chronological age using simple and accessible biomarkers - that was confronted by the scientists of ITMO University.
To create an algorithm predicting biomarker age, scientists used deep learning technologies. Deep neural networks (deep neural networks, DNN) were chosen because, according to scientists, they monitor hidden characteristics most effectively and can be trained on complex examples of multidimensional data.
Using DNN allowed scientists to track the non-linear relationships between blood test data and the chronological age of the patient. Anonymous data from more than 120,000 blood tests of patients from Canada, South Korea and Eastern European countries were used as input data for neural networks. To evaluate the effectiveness of their approach, scientists first used three independent neural networks - one for each population.
After that, the scientists compared the results of their work with the work of a neural network trained in a mixed sample made up of data from three datasets. All models used 19 parameters of the standard blood test (albumin, glucose, hemoglobin, etc.), as well as information about the patient's field. The neural network, trained on a mixed sample, also analyzed the data of belonging to a particular population.
To assess the performance of the networks and track their predictive capabilities, the researchers used the National Health and Nutrition Screening Program (USA) datasets - retrospective laboratory data and demographic data for the period from 1996 to 2016.
Nationality turned out to be one of the most significant markers in age estimation - according to scientists, this may be due to "differences in the types of aging between different populations."
As the scientists supposed, the networks that studied only on one uniform dataset predicted the age of people from another population is much worse than the network that studied on combined data. In the first case, the error in the assessment was up to 10 years, while the final version of the neural network determined the age with an error of less than 6 years.
Scientists also showed that the networks that studied on one dataset predict that the age of people of the same nationality is no better than a neural network trained on “combined” data — another argument in favor of the latter.
As the scientists note, each of the components of the blood test separately is not an accurate biomarker. However, in the case of a complex analysis, the combination of these parameters makes it possible to determine the person’s age relatively accurately. A detailed analysis of the data obtained by researchers is given in a scientific article .
Another result of this work is an online service for assessing the age of Aging.AI . Anyone who has the latest blood test results can use it.
By the way, this is not the only research in the field of medicine, which is conducted with the participation of scientists and researchers from ITMO University. About what other developments at the junction of medicine and IT are conducted at the University, we told in this material .
Unlike other developments, this one is more versatile and accurate.
In the course of the study, scientists analyzed data sets of people of different nationalities and showed that such an approach has more predictive accuracy than developments based on data of people belonging to the same population.
')
Read more about this study in our material.
Photo Aging.AI
Determining the exact age: why it matters
The definition of biological age is one of the promising areas of research related to the study of aging. Information about a person’s biological age (which may not always coincide with the chronological, or, in other words, “passport age”) can be used in the future when prescribing drugs or deciding whether a person can be a donor.
British researchers working in this area say that the ability to assess a person’s biological age (and compare it with chronological) can change not only medicine, but other industries as well, such as insurance, where age is an important indicator of scoring in some cases.
A simpler, but also non-trivial task is to search for biomarkers that allow one to estimate the chronological age. Such biomarkers, for example, can be a useful tool for assessing the therapeutic efficacy of a drug or procedure aimed at “slowing down aging”. It was this task - the determination of chronological age using simple and accessible biomarkers - that was confronted by the scientists of ITMO University.
How the study was conducted
To create an algorithm predicting biomarker age, scientists used deep learning technologies. Deep neural networks (deep neural networks, DNN) were chosen because, according to scientists, they monitor hidden characteristics most effectively and can be trained on complex examples of multidimensional data.
Using DNN allowed scientists to track the non-linear relationships between blood test data and the chronological age of the patient. Anonymous data from more than 120,000 blood tests of patients from Canada, South Korea and Eastern European countries were used as input data for neural networks. To evaluate the effectiveness of their approach, scientists first used three independent neural networks - one for each population.
After that, the scientists compared the results of their work with the work of a neural network trained in a mixed sample made up of data from three datasets. All models used 19 parameters of the standard blood test (albumin, glucose, hemoglobin, etc.), as well as information about the patient's field. The neural network, trained on a mixed sample, also analyzed the data of belonging to a particular population.
“The distribution of blood biomarkers for age determination varies in each population. We have created a neural network that levels these differences and can be used for any populations.
At the same time, it behaves more stably and makes less mistakes. We also identified key biomarkers that most affect aging. Among them are albumin and glucose. These data are consistent with what was previously known to change these indicators with aging. "
- Kirill Kochetov , one of the developers of the algorithm
To assess the performance of the networks and track their predictive capabilities, the researchers used the National Health and Nutrition Screening Program (USA) datasets - retrospective laboratory data and demographic data for the period from 1996 to 2016.
Research results
Nationality turned out to be one of the most significant markers in age estimation - according to scientists, this may be due to "differences in the types of aging between different populations."
As the scientists supposed, the networks that studied only on one uniform dataset predicted the age of people from another population is much worse than the network that studied on combined data. In the first case, the error in the assessment was up to 10 years, while the final version of the neural network determined the age with an error of less than 6 years.
Scientists also showed that the networks that studied on one dataset predict that the age of people of the same nationality is no better than a neural network trained on “combined” data — another argument in favor of the latter.
As the scientists note, each of the components of the blood test separately is not an accurate biomarker. However, in the case of a complex analysis, the combination of these parameters makes it possible to determine the person’s age relatively accurately. A detailed analysis of the data obtained by researchers is given in a scientific article .
Another result of this work is an online service for assessing the age of Aging.AI . Anyone who has the latest blood test results can use it.
By the way, this is not the only research in the field of medicine, which is conducted with the participation of scientists and researchers from ITMO University. About what other developments at the junction of medicine and IT are conducted at the University, we told in this material .
Source: https://habr.com/ru/post/350624/
All Articles