MyFiziq Limited (ASX: MYQ) claims to have a made a breakthrough in its development of the world’s first artificial intelligence and machine learning models that are able to mimic an individual’s medical images pertaining to body composition (including body fat percentage) via mobile device image capture.
The company says that while the predicated medical images are not a 100% replacement of an actual medical scan such asDual Energy X-ray Absorptiometry (DEXA), they are highly correlated and representative of the actual DEXA scans performed on the medical imaging machine.
According to MyFiziq, preliminary analysis of its developed models has shown that the predicated tissue and body fat medical images that we are able to generate have an average correlation of 95% and up to 97% in subjects with BMI of 35 or above, when compared to actual DEXA images, when ideal conditions are met.
Importantly, the relative body composition distribution is very promising, especially when a user wishes to track their fat or tissue distribution across their body at a single point in time. What does this capability mean to a user? It means that a user will not only be given their total body fat percentage, but also where that body fat is distributed across the individuals body at any point in time.
This breakthrough is the result of MyFiziq’s world class machine learning team and researchers in the area of computer vision, sport science, exercise, health and fitness, and its unique global collection of datasets of human imagery and medical scans obtained over the past 3 years. This unique data set and image capturing system puts MyFiziq at the forefront with respect to other competitors attempting to develop similar capabilities to the MyFiziq patented technology.
The example images depict the actual outputs of the machine learning models. As seen in the first image, MyFiziq’s state-of-the-art technology can extract 3D-like features just from the user’s standard mobile device images; however, the technology does not stop at this level because it further predicts the body composition imagery by [i] utilising the trained models based on our real-world unique data sets; and [ii], utilising the correlation of the complexed relationships between human body characteristics, biometrics, forensics and imagery and their DEXA-represented images.
This new capability has been developed and rigorously tested by the MyFiziq machine learning team lead by Dr. El-Sallam and Dr. Dhungel, who are the expertise behind the building and implementation of the new protocols. The team have been enhancing the internally collected machine learned DEXA imagery utilising thousands of individually collected medical images throughout the past 3 years.
This imagery was then trained within their data to identify and mimic the real-time DEXA imagery from this highly expensive medical imaging machine. The team then used real data captured using the MyFiziq technology to further enhance the existing model identification and relationships between the data and the human form.
CEOVlado Bosanac said the companyfollowed this with rigorous testing across thousands of participants where we have both DEXA imagery and on-device capture to draw a conclusion which has demonstrated a 95% accuracy between the MyFiziq DEXA mimicking and DEXA medical imagery and up to 97% in subjects with BMI of 35 or above.
This new capability has been part of our planned advance strategy for the last two years,” MrBosanac said.
“We accelerated the rollout due to the increase that we have seen in the Telemedicine and Telehealth industry due to the COVID 19 pandemic.
“We are addressing an immediate need posed by incoming enquiries and new partner opportunities. The Company, for want of a better explanation, is becoming a device-based health triage provider by allowing insurers, medical professionals, and healthcare providers to use an advanced tool, that demonstrates an individual’s risk markers with speed and convenience.
“MyFiziq assists in directional decision-making by health professionals at all levels. We expect to bring this additional capability to smartphone devices for our partners in early 2021.”