Techno Blender
Digitally Yours.

3D Body Scans Improve Predictive Models for Metabolic Syndrome | by lambert leong | Jul, 2022

0 57


3D Data for Health and Medicine

We use 3D body scans data to improve model prediction of metabolic disease

Summary TL:DR

  1. 3D data is high dimensional and can be used to build powerful models.
  2. Using 3D data (3D biomarkers) in health models helped improved predictions of metabolic syndrome status.
  3. We demonstrate another published example of using integrated discrimination improvement (IDI) and net reclassification index (NRI) to evaluate and compare two models and new biomarkers.
  4. Read the article

Three dimensional (3D) scanning technology has become cheaper, more accurate, and more accessible over the years. Gaming platforms such as the Xbox kinect contained cameras capable of capturing 3D body data and were highly accessible. The new iPhone 12 has a 3D scanning camera as well thus, demonstrating how the barrier to obtaining 3D data is diminishing fast. As a result, many have begun exploring powerful machine learning and deep learning models which incorporate fascinating technique such as graph and 3D convolutions. In this work we show how even simpler models can benefit from the proper utilization of 3D.

High resolution 3D scan taken in under 30 sec. Unlike this scan, research participants wear form fitting clothes to capture more accurate body shape (Image by author)

Healthcare and medical research has taken notice of 3D body scanning technology and many have begun to use it to monitor one’s health state. The idea is that you can look at a person and visually get a good intuition about the general health state of a person. It is also argued that this visual intuition is a more powerful health indicator than simple height and weight measurements as well as body mass index (BMI) resulting from those two measurements. In fact, my current research group currently has a grant from the National Institute of Health (NIH) to study how 2D and 3D body shape relate to metabolic disease, cancer, and overall health state. We have found that 3D scans are a powerful, comprehensive, and information dense piece of health data that is easy to acquire. In this work we looked at how body shape, captured by 3D scanners, helped us better predict if someone has metabolic syndrome.

Metabolic syndrome (MetS) is prevalent in one-third of the adults in the United States and is defined in the text below but the table may be easier to read. Essentially if three or more measurements meet the criteria, then that person has metabolic syndrome. Having MetS puts a person at a higher risk of heart disease and higher all-causes mortality. The good news is that a person has the ability to change their MetS status through modifying any of the measurements in the table above and the earlier the intervention the better the outcome.

Metabolic syndrome is defined using the 2005 National Cholesterol Education Program Adult Treatment Panel III (NCEP ATP III) guidelines as having ≥3 of the following: high waist circumference (as measured by manual anthropometry; ≥102 cm in men, ≥88 cm in women), elevated triglycerides (≥150 mg/dL), elevated blood pressure (≥130 mm Hg systolic or ≥85 mm Hg diastolic), elevated fasting glucose (≥100 mg/dL), and/or reduced HDL-C (< 40 mg/dL in men, <50 mg/dL in women). (Image by author)

Aside from waist circumference and blood pressure, the other measurements are more difficult to obtain and require a blood draw. As such, using accessible measures to build a model to predict MetS status would be beneficial. BMI is an easily accessible measurement that is often used; however, we believe it to be limiting. We also believe (or hypothesis) that 3D scans are more informative and they are just as accessible.

Image by author

The 3D scanning system produces a 3D mesh of a person and also reports all 3D anthropometric measurements which includes circumferences, volumes, and surface areas for the waist, arms, and legs to name a few. Circumference measures can be obtained manually with a tape measure however, it would require significantly more time to collect all regions. Therefore clinically, only waist circumference is often collected if at all. Several models were built using logistic regression and accessible patient data, other than blood markers, to predict MetS. The best model utilized BMI, age, sex, race, and 3D data to predict MetS with an AUC of 0.92. Our results demonstrate that MetS prediction models can be built using more accessible data that do not require a blood draw. We also show that BMI is fairly limiting and adding easy to acquire 3D data greatly increases model performance.

Below are 3D scans of Asian females around the same age with the same BMI. When we look at the scans however, we can see the differences in body shape that BMI does not account for. From the 3D data you can see differences in definition, fat distributions, muscle tone, and etc. Additionally, the female on the left has MetS while the female on the right does not and this makes sense, to a degree, when you look at the 3D scan. It ultimately is more informative than BMI alone.

Two participants of similar age and same body mass index (BMI) but one with metabolic syndrome (left) and one without (right). (Image by author)

The Integrated Discrimination Improvement (IDI) and Net Reclassification (NRI) are useful for comparing models and the use of new biomarkers. In this case we are looking at 3D biomarkers from 3D body scan data. I have a previous post/tutorial that demonstrates how to calculate it using python.

Image by author

We use these plots again to show how models which include 3D data improve on models using only BMI. Adding the 3D data resulted in integrated sensitivity and integrated specificity of 21.35 and 4.58, respectively indicating both an improvement to sensitivity and specificity. At the threshold, calculated using the Youden method, the net reclassification index for MetS and non-MetS was 9% and 8%, respectively. This means that with 3D data, the model was able to correctly identify 9% more individuals who have MetS that the model with BMI only missed. This also means that with 3D data, the model was able to correctly identify 8% more individuals without MetS than the model with only BMI that were incorrectly identified as having MetS.

Tutorial and more information on AUC, IDI, and NRI

Published IDI example

Grant and Study Data Request


3D Data for Health and Medicine

We use 3D body scans data to improve model prediction of metabolic disease

Summary TL:DR

  1. 3D data is high dimensional and can be used to build powerful models.
  2. Using 3D data (3D biomarkers) in health models helped improved predictions of metabolic syndrome status.
  3. We demonstrate another published example of using integrated discrimination improvement (IDI) and net reclassification index (NRI) to evaluate and compare two models and new biomarkers.
  4. Read the article

Three dimensional (3D) scanning technology has become cheaper, more accurate, and more accessible over the years. Gaming platforms such as the Xbox kinect contained cameras capable of capturing 3D body data and were highly accessible. The new iPhone 12 has a 3D scanning camera as well thus, demonstrating how the barrier to obtaining 3D data is diminishing fast. As a result, many have begun exploring powerful machine learning and deep learning models which incorporate fascinating technique such as graph and 3D convolutions. In this work we show how even simpler models can benefit from the proper utilization of 3D.

High resolution 3D scan taken in under 30 sec. Unlike this scan, research participants wear form fitting clothes to capture more accurate body shape (Image by author)

Healthcare and medical research has taken notice of 3D body scanning technology and many have begun to use it to monitor one’s health state. The idea is that you can look at a person and visually get a good intuition about the general health state of a person. It is also argued that this visual intuition is a more powerful health indicator than simple height and weight measurements as well as body mass index (BMI) resulting from those two measurements. In fact, my current research group currently has a grant from the National Institute of Health (NIH) to study how 2D and 3D body shape relate to metabolic disease, cancer, and overall health state. We have found that 3D scans are a powerful, comprehensive, and information dense piece of health data that is easy to acquire. In this work we looked at how body shape, captured by 3D scanners, helped us better predict if someone has metabolic syndrome.

Metabolic syndrome (MetS) is prevalent in one-third of the adults in the United States and is defined in the text below but the table may be easier to read. Essentially if three or more measurements meet the criteria, then that person has metabolic syndrome. Having MetS puts a person at a higher risk of heart disease and higher all-causes mortality. The good news is that a person has the ability to change their MetS status through modifying any of the measurements in the table above and the earlier the intervention the better the outcome.

Metabolic syndrome is defined using the 2005 National Cholesterol Education Program Adult Treatment Panel III (NCEP ATP III) guidelines as having ≥3 of the following: high waist circumference (as measured by manual anthropometry; ≥102 cm in men, ≥88 cm in women), elevated triglycerides (≥150 mg/dL), elevated blood pressure (≥130 mm Hg systolic or ≥85 mm Hg diastolic), elevated fasting glucose (≥100 mg/dL), and/or reduced HDL-C (< 40 mg/dL in men, <50 mg/dL in women). (Image by author)

Aside from waist circumference and blood pressure, the other measurements are more difficult to obtain and require a blood draw. As such, using accessible measures to build a model to predict MetS status would be beneficial. BMI is an easily accessible measurement that is often used; however, we believe it to be limiting. We also believe (or hypothesis) that 3D scans are more informative and they are just as accessible.

Image by author

The 3D scanning system produces a 3D mesh of a person and also reports all 3D anthropometric measurements which includes circumferences, volumes, and surface areas for the waist, arms, and legs to name a few. Circumference measures can be obtained manually with a tape measure however, it would require significantly more time to collect all regions. Therefore clinically, only waist circumference is often collected if at all. Several models were built using logistic regression and accessible patient data, other than blood markers, to predict MetS. The best model utilized BMI, age, sex, race, and 3D data to predict MetS with an AUC of 0.92. Our results demonstrate that MetS prediction models can be built using more accessible data that do not require a blood draw. We also show that BMI is fairly limiting and adding easy to acquire 3D data greatly increases model performance.

Below are 3D scans of Asian females around the same age with the same BMI. When we look at the scans however, we can see the differences in body shape that BMI does not account for. From the 3D data you can see differences in definition, fat distributions, muscle tone, and etc. Additionally, the female on the left has MetS while the female on the right does not and this makes sense, to a degree, when you look at the 3D scan. It ultimately is more informative than BMI alone.

Two participants of similar age and same body mass index (BMI) but one with metabolic syndrome (left) and one without (right). (Image by author)

The Integrated Discrimination Improvement (IDI) and Net Reclassification (NRI) are useful for comparing models and the use of new biomarkers. In this case we are looking at 3D biomarkers from 3D body scan data. I have a previous post/tutorial that demonstrates how to calculate it using python.

Image by author

We use these plots again to show how models which include 3D data improve on models using only BMI. Adding the 3D data resulted in integrated sensitivity and integrated specificity of 21.35 and 4.58, respectively indicating both an improvement to sensitivity and specificity. At the threshold, calculated using the Youden method, the net reclassification index for MetS and non-MetS was 9% and 8%, respectively. This means that with 3D data, the model was able to correctly identify 9% more individuals who have MetS that the model with BMI only missed. This also means that with 3D data, the model was able to correctly identify 8% more individuals without MetS than the model with only BMI that were incorrectly identified as having MetS.

Tutorial and more information on AUC, IDI, and NRI

Published IDI example

Grant and Study Data Request

FOLLOW US ON GOOGLE NEWS

Read original article here

Denial of responsibility! Techno Blender is an automatic aggregator of the all world’s media. In each content, the hyperlink to the primary source is specified. All trademarks belong to their rightful owners, all materials to their authors. If you are the owner of the content and do not want us to publish your materials, please contact us by email – [email protected]. The content will be deleted within 24 hours.

Leave a comment