INTRODUCTION
How can we improve the donor selection to increase the resemblance between receptors and their offspring?
It’s necessary to implement new algorithms including biometry distance measurements (Fig.1.1) across family generations to improve donor selection according to the facial resemblance.
What is Fenomatch?
Fenomatch is a facial matching tool that helps doctors find the right donor for each parent.
The Artificial Intelligence algorithm compares the parent’s face with the faces of the potential donors, focusing on phenotypic traits- those which are passed on to our children via DNA.
Your doctor can choose the donor with the greatest facial resemblance.
The platform also checks other phenotypic traits like eye colour, hair colour, ethnicity, skin tone, etc. as well as generic compatibility and blood type.
WHAT CAN WE DO WITH FENOMATCH?
Filter donors by their facial resemblance to the patient
Use AI to compare over 12,000 biometric data points
Check genetic compability
Check donor/patient compability with just one click
Minimise the risk of human error
Automatically verifying traits such as ethnicity and gender, and providing traceability
Provide peace of mind to the patient
A certificate verifies the use of the latest techology in the donor selection process
Save time and improve efficiency
Give your clinical team objective criteria to make the right decision
Find the right donor
Work with your own gamete bank, or connect with others
Fenomatch in numbers
We works over
Outtechnology is used on
Algorith with over
DATA SECURITY
This platform is built on 100% secure technology.
METHODS
A total of 864 subjects (age 18-60) from 108 families (Fig.1.2) were included with their participant agreement (73% training – 27% validation).
All images correspond to 21-50 years old subjects, with a gender distribution 1:1.
80 Gb RAM and 60 computing processors were used intensively.
Artificial Neural Networks methodologies were finally used to find a predictable pattern.
RESULTS
The new algorithm is stable and can reproduce results when iterating validation data.
The new prediction model based on biometric distances results is 86,66% (σ2 47.88) if we require all images must be in the right order to get a positive result.
The accuracy results improve performance up to 96% when the right order is just required in the 5 images with more resemblance.
Comparison between images is independent of the subject gender providing same accuracy results for gender-crossed images as for same-gender.
As final extra testing, when the test was made by the computational algorithm and also by a human team (10 individuals with a majority vote criteria), humans results are 98% in line with the new algorithm.
FENOMATCH ALGORITHM VS FACIAL RECOGNITION
When doing the same testing with the two most used existing open python scripts based on facial recognition average results provide 56.88% of accuracy (σ2 638.98) in ordering subjects right.
When comparing gender-crossed images test shows just a 23.38 (σ2 89.23) of accuracy.
This comparison result makes sense since facial recognition techniques are not intended to compare genealogical similarities, but only have the purpose of biometric identification of a person.
INTRODUCTION
We can conclude that Fenomatch new algorithm is valid as it has great accuracy in its results and it has widely improved results in the current state of the art.
The Fenomatch algorithm should be used as a decision support application that helps the medical team decide on a donor from among those who already meet the medical and genetic criteria in order to guarantee to find the donor that most resembles the receptor’s family.
Other approaches based on facial recognition techniques does not show enough accuracy to be used for donor selection purposes.
