The use of Machine Learning to assist in the selection of the "best"
embryos for transfer in Human In Vitro Fertilisation
Despite many advances in Human In Vitro Fertilisation (IVF) technology,
success rates remain disappointingly low. In the UK only 1 in 7 infertile
couples go home with a child. Although both maternal and paternal factors
are involved, the quality of the embryos transferred to the mother is of
paramount importance. This study aims to improve IVF success rates by developing
rules to assist in the selection of the "best" embryos for transfer. These
rules will be obtained by using computerised "induction" techniques developed
in Machine Learning research.
IVF treatment typically involves the collection of 6-8 eggs from the
ovaries of the woman, which are then fertilised in vitro with partner or
donor sperm, resulting in 6-7 embryos. The law however, only permits 3
to be transferred to the woman's uterus. Currently, in Oxford, as in other
IVF Units, these 3 embryos are selected on the basis of subjective criteria
involving approximately 6 features of the embryo's morphology, and its
follicle and oocyte of origin. However, more than 60 features, describing
each embryo, oocyte, follicle and sperm sample, are available and recorded
in our extensive laboratory database. Clearly, it is impossible for the
embryologist to simultaneously assess all of these features, but state-of-the-art
computer-aided induction techniques can. We have therefore developed a
novel and exciting collaboration between the Nuffield Department of Obstetrics
and Gynaecology (Dr I.L. Sargent) and the Oxford
University Computing Laboratory to use Machine Learning to develop
rules for embryo selection.
In this approach, the learning program searches for discriminatory patterns
by studying all of the features of specific batches of embryos from our
database which either did, or did not, result in a live child. These patterns
are then expressed in the form of "if-then" rules. Having established these
rules retrospectively they can then be used prospectively in the IVF Clinic
to identify the embryos which have the greatest potential to develop into
a live child.
This proposal is motivated by promising initial results obtained by
Dr R. Saith of this laboratory. Working in collaboration with a member
of Dr Muggleton's group, in a study of 81 embryos Dr Saith has developed
a limited set of "embryo selection" rules which could discriminate between
2 groups of "good" and "bad" embryos. The successful completion of this
pilot study will permit the development of more sophisticated rules which
take into account embryo-related as well as maternal and paternal factors.
These could have a major impact on IVF success rates world wide.
Dr. Ian Sargent, Nuffield Department of Obstetrics and Gynaecology.
Ruhi Saith, Nuffield Department of Obstetrics and Gynaecology.