Assisted fertilisation processes involve striking the perfect balance between a range of complex factors – all of which play a critical role in getting pregnant, including ovulation cycles, hormone levels, lifestyle factors and more.  At a number of stages throughout assisted reproductive technology (ART) processes, decisions are made by clinicians, for example, in analysing images of oocytes or selecting sperm and eggs.  These clinical decisions necessarily impact on the success or failure of this process.  While the expertise and knowledge of fertility doctors remains unparalleled, the introduction of artificial intelligence (AI) into ART processes can maximise the success rate and efficiency of fertility treatments, incorporating AI’s more objective “intelligence” into the decision-making process.

AI in gamete and embryo selection

The selection of good quality gametes (e.g. eggs or sperm) is of primary importance in securing a successful pregnancy.  The quality of female gametes may be impacted by follicle size, oocyte morphology and cytoplasmic features, whilst concentration, motility and morphology are known to influence the quality of sperm.  The use of AI can remove the subjectivity prevalent in human selection of gametes as AI can objectively assess gamete health based on quality (though, of course, it is well-known that issues can arise if an AI system itself is biased, for example, due to the use of biased training data).

Embryologists are often tasked with selecting sperm for intracytoplasmic sperm injection.  AI may assist with further developing the assessment criteria used to ensure high quality sperm are selected by identifying new factors that influence the quality of sperm (e.g. direction of movement, swimming patterns).  In this way, AI may assist with identifying the ideal sperm-egg combination to optimise the success rate of fertilisation.

Similarly, AI can be used in embryo selection to successfully bring about pregnancy. AI has the ability to detect and assess components known to indicate high quality embryos and flag complex patterns almost unidentifiable to the human eye by studying traits of previous embryos that have progressed to successful pregnancies. Through routinely generated images or time-lapse videos, AI can grade, assess, and rank embryos to ensure the optimal embryos are selected for transfer. Thus, AI enhances the accuracy of grading and the prediction of blastocyst formation and pregnancy, and reduces the risk of intra- and inter-operator variability.  

The issues of AI in fertility 

Fertility-focussed AI and machine learning technologies are far from perfect.  One major concern is the use of uninterpretable or “black-box” models that are either too complex for human comprehension or are proprietary (and therefore, only visible and interpretable by the owners of the system).  Several research studies have applauded the outstanding performance of AI in selecting embryos, noting the admirable accuracy of these machine learning models.  However, a number of these studies train and test software based on incredibly poor-quality embryos or broad characterisations of “good” or “poor” quality embryos, largely defeating the purpose of using AI in making decisions such as the selection of the highest quality embryo for implantation.  The reality is that most embryologists would, at the outset, discard any “poor” embryos from the selection pool, only deciding amongst the highest quality embryos.

The purpose behind integrating AI into fertility treatments is to increase the level of information and knowledge available to doctors in decision-making, where human intelligence and comprehension is limited.  There is no motivation to use AI that simply provides information that is already understood by doctors.  Until AI and machine learning software develop and train themselves to a point where they can identify high quality gametes and embryos in order to progress to a successful pregnancy there may be little use for such technology in clinical decision-making.

Additionally, there are broader ethical concerns in relation to the use of AI and machine learning technology, including potential overreliance and complacency as a result of a model’s automated decision-making.  To some extent, this may be mediated by regulatory oversight.  For example, under Australian law, medical devices (which includes software as a medical device (SaaMD)) that make all the necessary decisions (e.g. for diagnosis or screening) are subject to the highest level of scrutiny by the Therapeutic Goods Administration (TGA) as Class III medical devices.  However, where software only provides information to a relevant health professional to assist them, and the health professional is responsible for the final clinical decision-making, a lower classification (and thus less regulatory oversight) is assigned.

Moreover, ethical concerns have been raised about the potential rise of “designer babies”, which feeds into the debate about whether patients’ own preferences should be inserted into AI algorithms for embryo selection.  The question then becomes: will AI prioritise or favour those values (e.g. sex, intelligence, physical characteristics) in its decision-making over and above finding the most high-quality embryo (i.e. the one with the most chance of resulting in a successful pregnancy) for implantation.  These ethical issues are only going to become more pressing with the advent of new technologies such as in-vitro gametogenesis (IVG), which is a process whereby embryos may be grown in a lab by reprograming adult cells (e.g. skin cells) to become sperm and egg cells.

Despite the broader ethical dilemmas surrounding the technology, there is obviously merit to the integration of AI in the fertility space, and its prevalence and application alongside human intelligence in the decision-making process will only continue to grow as new technologies develop.

Questions facing inventors using AI in ART

An important question facing inventors and companies working in this space is whether these new inventions are likely to be patentable subject matter in Australia, especially in light of the Full Court’s decision in Ariosa Diagnostics, Inc v Sequenom, Inc [2021] FCAFC 101 (Ariosa) (which was discussed in our previous article) and section 18(2) of the Patents Act 1990 (Cth) (Patents Act), and also possibly the High Court’s decision in Aristocrat Technologies Australia Pty Ltd v Commissioner of Patents [2022] HCA 29 (Aristocrat).

The question for determination in Ariosa was whether Ariosa Diagnostics Inc and its licensees’ “Harmony Test” had infringed certain claims in Sequenom’s patent.  Integral to the Full Court’s deliberations was whether the Harmony Test in itself was patentable subject matter (the Harmony Test used a non-invasive prenatal diagnosis technique involving detection of the presence of foetal nucleic acid to determine the risk of genetic malformations or chromosomal abnormalities).  The Full Court recognised that the invention was a patentable invention (i.e. was a “manner of manufacture” within the meaning of section 18(1)(a) of the Patents Act): the claims were not, as a matter of substance, directed to genetic information, but to a method involving the practical application of a means for identifying and discriminating between maternal and foetal nucleic acid.  Although foetal nucleic acid occurs in nature, the substance of the invention was not the nucleic acid itself, but the identification of that particular nucleic acid as a part of a method.  

While section 18(2) of the Patents Act excludes from patentability “human beings and the biological processes for their generation”, which is interpreted by the Australian patent office to cover embryos and processes for generating or culturing human embryos, based on the Full Court’s decision in Ariosa, it is likely that processes for gamete and embryo selection as part of fertility treatments will be patentable subject matter (so long as the fertilisation and the implantation of the embryo are not also claimed).  This is because, whilst gametes and embryos are naturally occurring, a process to identify and discriminate between high and low quality gametes and embryos yields a useful outcome (e.g. an increased chance of a successful pregnancy), and is not itself a process for generating or culturing an embryo (which would be contrary to s 18(2) of the Patents Act).  However, inventors will need to consider the precise metes and bounds of their invention and the claims they intend to include in any patent application to assess whether or not that invention will be patentable in light of Ariosa and s 18(2) of the Patents Act.

Inventors and companies looking to claim the use of AI and machine learning as part of such a process should also consider whether they have claimed their invention in a manner that is patentable subject matter in view of the High Court’s split decision in Aristocrat, which failed to clarify the law in Australia regarding the patentability of computer-implemented inventions.  While the High Court agreed that in order for a computer-implemented invention to constitute a “manner of manufacture”, it must be more than a mere scheme or abstract idea that happens to be implemented using a computer, the High Court was evenly split in how the invention in question should be characterised and, as a result, whether the invention was patentable.

On the one hand, Chief Justice Kiefel, and Justices Gageler and Keane held that absent a claim to a variation of standard computer technology to implement or accommodate the idea (a game), the invention was nothing more than an abstract idea and was thus not patentable subject matter.  On the other hand, Justices Gordon, Edelman and Steward found that there was a patentable invention because the invention was an electronic gaming machine that was altered by the inclusion of the games and configurable symbols - the invention was more than the simple incorporation of an idea of a game into a generic gaming machine.  In light of the High Court’s split decision, the Full Federal Court’s decision was affirmed. 

As a result, the state of the law in Australia as to the true test to be used to ascertain whether a computer-implemented invention is a manner of manufacture remains uncertain, though IP Australia has stated that an invention will not be considered a “manner of manufacture” if it is merely directed towards the implementation of an otherwise unpatentable idea in conventional computer technology.  Thus, depending on how AI or machine learning are claimed as part of claims to processes directed to the use of ART, it is possible that while some claims may be determined to claim patentable subject matter, others may not.