Introduction
‘Big data’ analytics solutions are key to what the future has in store for ART professionals. These solutions are already here and the specialised technology platforms that use such solutions are currently operational or under development. The main advantage of big data analytics is that they can accurately predict individual patient success rates based on specific circumstances. In order to be effective, these platforms require an effective database containing thousands of patient profiles with very detailed information on medical history and fertility examinations as well as outcome data. Platforms use these data to find similarities and to draw conclusions regarding potential patient outcomes, thus estimating individual success rates.
The large amounts of data available for each type of treatment and fertility problem create an obstacle that platform developers have to overcome. However, the collection of data from a variety of sources/ clinics/countries is vital in order to accurately predict the chances of success.
If big data analytics is paving the way to the future, Artificial Intelligence (AI) solutions are the future. AI builds on big data analytics to further enhance technology platforms that estimate IVF success rates. The ability of ‘machine learning’ and the identification of data patterns that are accurately interpreted by AI systems to reach conclusions potentially beyond the realms of human intelligence, are major challenges when it comes to using technology to estimate success.
AI can help us not only to estimate success rates but also to recommend treatments, methods and protocols to maximise the chances of success. It can also recommend the best locations/clinics for optimal individual outcomes.
AI systems are currently at the research and development stage. They can impact three main areas of fertility. The first is outcome prediction, where the further development of AI algorithms will allow systems to estimate outcome percentages more accurately, as discussed above. The second is the development of clinical decision systems capable of providing a better patient service, selecting the appropriate patient protocol, and evaluating suitable types of treatment and laboratory methods to optimise results. Finally, AI systems can be used to assess cost-effectiveness and reduce unnecessary expenses from failed cycles.
Robotic surgery is one of the areas in which AI has been widely applied and is still under research. It has not yet been applied to fertility but maybe this is the next step in assisted reproduction.
Different uses of Artificial Intelligence in ART
The lack of automation and the vast array of clinical presentations and treatment options in ART lead to considerable inter-user variability in terms of healthcare provision. This variability can lead to poor outcomes and may be confusing for patients. Artificial Intelligence can be used to reduce this variability by learning from vast amounts of clinical, demographic, pathological, imaging and laboratory data, and making connections and recommendations to guide health care decisions. Automating and streamlining the entire process should reduce the overhead costs for fertility practices and increase patient access (1).
Artificial Intelligence applications are emerging in different areas of ART (3):
- Triage, screening, and diagnosis
- Prediction of outcomes
- Treatment personalisation and monitoring
- Image interpretation
Use of AI in triage, screening and diagnosis
AI is used to efficiently interpret large health datasets in the context of clinical triage, screening, and diagnostics, These AI systems are trained on external health data that have usually been interpreted by humans and that have been minimally processed before exposure to the AI system, for example, clinical images that have been labelled and interpreted by a human expert. The AI system learns to execute the interpretation task on new health data of the same type, which in clinical diagnostics is often the identification or forecasting of a disease state (4).
It is particularly useful for clinicians with triaging patients who might need fertility treatment sooner, when screening for potential problems or for investigating factors that can affect reproductive health. It could even be used to diagnose specific conditions such as endometriosis, polycystic ovary syndrome or diminished ovarian reserve.
Use of AI in prediction of outcomes
Prediction of outcomes and estimating success in IVF is a complicated task that requires new technological solutions if it is to be done effectively. Traditionally, success rates are estimated by medical experts based on the medical history and current examination findings of the patient in question as well as on the expert’s experience and knowhow. However, this can never be accurate and can vary depending on the clarity of mind of the expert at the point of consultation, factors such as tiredness, ability to remember data from past cases, and potential to analyse and evaluate statistical data, etc. Current developments in technology are providing us with important tools to eliminate human errors and miscalculations and to provide accurate estimated success rates easily and quickly. Patients can already identify clinic websites in different countries that offer simple solutions. This quickly gives them an idea of their chances of success based on broad categories such as age group, type of treatment and main fertility problem, etc. These solutions are widely available and constitute the first step in showing what technology can offer in terms of accurately estimated IVF success rates. One problem with these solutions is that they are too broad and generalise findings without examining individual patient specifics.
Some research has been conducted in this area. Gil et al. looked at utilising various AI networks to analyse the association between environmental factors and/ or lifestyle habits and the potential impact on semen quality (5). Some of the variables assessed included smoking, alcohol consumption and body mass index (BMI). The data obtained displayed a high predictive accuracy (~ 86%) for sperm concentration, and (73–76%) for motility. Another example is presented in a study by Candemir et al. in the form of an alternative algorithmic model to predict semen quality based on a similar questionnaire with additional variables including season of analysis and history of genitourinary trauma (5). A radial basis function neural network was used in this model, and success rates of up to 90% were reported in estimating semen quality. This was deemed to be the most accurate for predicting semen quality compared to previous models (5). El- Shafeiy et al. have also demonstrated further optimisation of an ANN for predicting fertility quality by coupling an additional optimisation algorithm, termed the Sperm Whale Optimisation algorithm (5). Predictive models for semen quality could be used as the initial step in screening men who may need an infertility evaluation. Earlier identification of men potentially presenting sub-fertility by recommending a semen analysis would lead to earlier intervention for couples desiring pregnancy.
In the fertility treatment context, a different study reviewed 95, 868 medical records and created a dynamic grading system that considered seven indicators, namely age, body mass index, follicle-stimulating hormone level, antral follicle count, anti-mullerian hormone level, number of oocytes and endometrial thickness to predict treatment outcomes (2). The system graded the patient’s infertility on 5 levels ranging from A to E with A corresponding to a pregnancy rate of 53.82% and E to a 0.90% pregnancy rate. The cross-validation results confirmed system stability of 95.94% (95% CI, 95.14% – 96.74%) (2). The authors point out that this machine learning–derived algorithm may assist clinicians in making an efficient and accurate initial judgment on the condition of infertility patients (2).
Use of AI in treatment personalisation and monitoring
Different patients respond differently to fertility drugs and treatment protocols. Personalised treatment therefore has significant potential to improve the outcomes that matter to patients. However, many factors need to be considered when attempting to personalise treatment protocols including age, body mass index (BMI), hormone levels and ovarian reserve capacity, to name but a few. This can prove challenging (2). Artificial intelligence and machine learning can be used to automate the complex task of analysing all of the individual factors, to compare against a benchmark and make personalised treatment protocol recommendations. Similar algorithms can be used to monitor the patient’s response during treatment, assessing, for example, the risk of OHSS or poor ovarian response.
Use of AI in imaging interpretation
Imaging interpretation is potentially the most advanced area of Artificial Intelligence. However, although obstetric and gynaecological ultrasound scans are two of the most widely performed imaging studies, AI has had little impact on this field so far. Nevertheless, there is huge potential for AI to assist in repetitive ultrasound tasks, such as automatically identifying good-quality acquisitions and providing instant quality assurance (3). In this area, for each ultrasound task, there are several image acquisition and analysis capabilities that can be met by an AI application, including classification (‘what objects are present in this image?’), segmentation (‘where are the organ boundaries?’), navigation (‘how can I acquire the optimal image?’), quality assessment (‘is this image fit for purpose to make a diagnosis?’) and diagnosis (‘what is wrong with the imaged object?’) (3). In obstetric and gynaecological ultrasound, promising workload-changing advancements include automatic detection of standard planes and quality assurance in foetal ultrasound, detection of endometrial thickness in gynaecology and automatic classification of ovarian cysts.
Perhaps the most important application of Artificial Intelligence is in the IVF lab. Historically, embryo selection for transfer has been based on human experience and expertise. The skills and seniority of the embryologist would play a vital role in selecting the best embryo for transfer, thus speeding up the process and eliminating any unnecessary, unsuccessful attempts. This problem has been addressed to a certain degree by expensive infrastructure, such as time lapse incubators that provide extra tools and scope for embryologists. However, this remains a personal human decision.
AI offers a system that learns from embryo development patterns and the implantation success of those embryos. In this way, the AI system would be able to effectively grade embryos according to real success potentials and thus standardise the embryo selection process and eliminate human error. Furthermore, AI systems operate algorithms that compare input data with output data and develop a process known as ‘Machine Learning’. Provided that good quality data are available in sufficient quantities, this process allows Artificial Intelligence to learn to improve itself, thereby providing more accurate predictions over time.
The application of AI solutions is available not only at the embryo selection stage but also one step earlier, i.e. the selection of eggs and sperm. Although the selection of eggs for fertilisation is not such an issue as all available eggs are usually fertilised, the selection of sperm is of vital importance. Research is currently underway and solutions are being tested on how to select the best available sperm using AI algorithms. This is particularly important in cases of sperm samples with low normal morphology, high fragmentation and altered DNA.
In conclusion, artificial intelligence is currently being applied in several areas to improve clinical decision-making. Further research and development is ongoing so that these systems can become more proficient, improving success rates and reducing errors as well as generating cheaper, faster and more accessible results.
Resources
- https://www.fertstertdialog.com/posts/ai-will-revolutionize-assisted-reproductive-technology-if-we-work-together
- https://jamanetwork.com/journals/jamanetworkopen/ fullarticle/2772681
- https://obgyn.onlinelibrary.wiley.com/doi/10.1002/ uog.22122 https://genomemedicine.biomedcentral.com/ articles/10.1186/s13073-019-0689-8
- http://website60s.com/upload/files/5-artificial-intelligence-in-reproductive-urology.pdf
- https://www.ncbi.nlm.nih.gov/pmc/articles/ PMC6733338/
- https://healthmatters.nyp.org/behind-the-latest-advance-in-ivf-treatment/
- https://link.springer.com/article/10.1007/s10815-020- 01881-9