To increase the success of IVF, hire an artificial embryologist
One
in six couples experience infertility, which is a global health problem.
Unfortunately, universal healthcare frequently does not cover treatment, and
for many, the cost is too high. Even in situations where these financial
obstacles can be overcome, success rates have stalled at about 33%, which means
that just one-third of all treatments result in fertilisation.
Deirdre
Zander-Fox, regional scientific director of Monash IVF and a knowledgeable
clinician, stated that "success rates are appearing to plateau,
which is likely due to a large number of factors, including increased maternal
age at the time of first pregnancy, which has significant impacts on the
quality of eggs, contribution of increasing patient BMI (both male and female),
as well as challenges in the technology space".
The
inability to distinguish "good quality" sperm, a crucial deciding
element for usage in the treatment cycle, contributes to the issue. "At
present, scientists choose sperm and embryos depending on how they appear under
a microscope at defined discrete timepoints," said Zander-Fox. It is hoped
that new research and scientific advancements would increase the success rates
of ART in conceiving and giving birth to live children.
In
order to choose a candidate for injection into the oocyte during assisted
reproduction cycles, the shape of the sperm head has been demonstrated to
correlate with other sperm quality parameters. Despite improvements in tools
and technology, an embryologist's judgement ultimately governs the matter. The
fact that this is a highly manual and subjective procedure and that there are
discrepancies across clinics that teach different identification techniques as
well as people who use their own tactics and shortcuts results in relatively
low success rates.
According
to Reza Nosrati, a researcher at Monash University, "I think embryologists
are doing their best and taking the hit for this, but there is not enough
technology available to them that can allow them to standardise and automate
this procedure."
Nosrati
suggests that a substitute for a human embryologist could be an artificial one,
automating and standardising the process entirely. Since deep learning
algorithms can self-train and recognise complicated patterns with little
subjectivity, they could simplify the classification of sperm. These
"virtual embryologists" could give sperm analysis and selection a
global, unchanging frame of reference.
To
do this, Nosrati and his group at Monash University developed an ensemble
learning technique for identifying microscope images of human sperm cells as an
embryologist would. The journal Advanced Intelligent Systems has published a
report on its findings.
Zander-Fox,
who was not involved in the study, expressed his interest in the application of
AI and machine learning to evaluate sperm. We are aware of the issue with the
current inter- and intra-operator variability in this area, as well as the
limitations of the human sight. The pace of the processes would increase, and
hopefully, the quality of the embryos we produce for treatment as well, if
there was a method to apply AI and machine learning to help in the selection of
the best sperm to use for insemination.
Their
model's use of an "ensembled approach," which employs many learning
algorithms in place of just one to improve predicted performance, distinguishes
it from others.
A
group of online embryologists
Four
independent picture classification models, known as convolutional neural
networks (CNNs), were trained separately in the study to create four virtual
embryologists who were experts in the classification of sperm heads.
Because
each model has a unique internal structure and the training procedure is
random, each model learns a unique link between the cell picture and the shape
categorization. As a result, they gain a slightly different comprehension of
the sperm morphology and slightly different positive qualities of the sperm
cells.
The
purpose of our effort, according to Nosrati, is to transform the current
platform into an automated system that can mark sperm cells in real-time and is
connected to the system utilised by embryologists during an ART cycle.
Hopefully, the embryologist can choose a better sperm for injection by choosing
one of the sperm cells that have a better likelihood of fertilising the egg by
using the real-time platform to highlight them.
This
approach, which classifies each cell using a collection of algorithms, is comparable
to a team of experienced physicians, each bringing their own set of
experiences, viewpoints, and views, working together to categorise each cell. A
meta-model, which serves as a supervisor for the four base models by ingesting
highly informed and distilled information from many sources and making an
executive decision, can then be utilised to intelligently leverage the
findings.
Although
a positive move, there are still challenges to be solved. The platform needs to
be improved in order to label and evaluate the entire sperm rather than just
the head, as well as to take other crucial factors like sperm motility and DNA
integrity into account, according to Nosrati.
This
work indicates a chance to better leverage machine learning's capabilities to
provide more effective clinical treatments. Experts still remain a viable
option, though.
Although
AI and machine learning will play a big part in medicine in the future, there
are concerns about the lack of transparency as to how the AI/machine learning
model was created and what inputs were used, which needs to be addressed to
ensure that medical and scientific experts trust in these systems," said
Zander-Fox. Also, to be mentioned is the fact that these tools were created to
support scientists and clinicians in their decision-making and should not be
utilised in place of the knowledge that scientific medical experts contribute.
We cannot guarantee the best patient outcomes without a partnership between
AI/machine learning and professionals.
According
to the research team, their findings can be used in any field where machine
learning is used, not simply sperm categorization or even picture recognition
tasks. The group must first solidify their existing platform, though, before
that.
In
three to five years, a beta prototype should be ready, according to Nosrati.
"In partnership with fertility clinics, we have already begun testing this
platform. We require the best training data set in order to obtain the best
algorithm. The approach should be guided not just by measurements of sperm
quality but also by the actual result: a live birth.