Introduction

Facilitating a pregnancy in a laboratory was one of the most impactful medical breakthroughs of the 20th century. The remarkable discovery of in-vitro fertilization (IVF) gave millions of people with an inability to naturally reproduce a second chance at having children. In the 21st century, scientists are looking to advance IVF even further through the implementation of new technologies. Some may ask, why fix something that isn’t broken? The answer is to make IVF treatment smarter and more effective. With the recent boom of artificial intelligence (AI) and the power of machine learning, the limits of possibilities and human error are becoming more and more diminishable. AI is revolutionizing the field of medicine, pushing the boundaries of innovation and transforming the way humans are treated. Fusion between IVF and AI will transform the approach to medical treatment for generations to come. 

What is IVF?

In-vitro fertilization (IVF) is a form of treatment for infertility designed to help individuals unable to organically conceive have healthy pregnancies. Infertility is classified as a disease of the male or female reproductive system which can lead to failure in achieving a pregnancy after 12 months of natural intercourse. According to the World Health Organization (WHO), 1 in 6 people—around 17.5% of the adult population—globally are affected by infertility. IVF provides a life-changing option for people to have children after diagnosis of infertility. In IVF, medical professionals called embryologists specialize in carrying out the medical process. An embryologist takes care of the embryos from the time of egg retrieval until the fertilized embryo is implanted into the mother’s uterus. A key part of an embryologist’s job is to grade and observe the developing embryos in order to identify the ones most likely to result in a successful and healthy pregnancy. By performing laser biopsies to screen for genetic diseases and analyzing scans to interpret the features of the embryo, an embryologist’s judgment is crucial to determine the outcome of the IVF treatment. 

During IVF, mature eggs are collected from ovaries and manually fertilized by sperm in a lab setting. Using a complex procedure, sperm is injected into a precise location at a specific orientation of an egg by an embryologist. The pair are given a short amount of time to develop and then after extensive testing, one of the fertilized eggs is chosen. Based on the embryo’s maturity and prediction to be successful, the chosen embryo is implanted in the uterus. The embryo then develops in the uterus and becomes a pregnancy with time. One full cycle of IVF takes about two to three weeks. There are several causes for infertility, and IVF holds answers for many of them. IVF can be used to address various conditions: fallopian tube damage, ovulation disorders, endometriosis, uterine fibroids, and genetic sperm issues. For example, if a mother’s uterus is unable to carry a child, but both the mother and father have viable egg and sperm, respectively, IVF can be used on a surrogate pregnancy carrier. The biological genetic cells would come from both parents. Or, if the mother’s eggs are not viable, IVF can be carried out with a donor egg and the father’s sperm to be implanted in the surrogate’s uterus. IVF, a combination of many different procedural techniques, became a revolutionary technology used to help millions around the world.

Artificial Intelligence Development

Artificial Intelligence (AI) is beginning to change the way the world functions by opening up infinite possibilities in a variety of fields. AI aims to create machines and computational systems that can perform human tasks, oftentimes exceeding human capabilities. These systems can process large amounts of data, recognize relationships, and use these skills to perform tasks or respond to situations. In the application of machine learning in IVF treatment, a specific branch of AI, called Convolutional Neural Networks, are of the utmost importance.

Convolutional Neural Networks (CNNs) are a form of AI machine learning and the core of deep learning algorithms. Layers of nodes send messages to each other in data communication networks, creating the inherent intelligence of CNNs. CNNs specialize in feature extraction and identifying patterns within an image. There are typically three main layers within a CNN that give it the ability to perform more efficiently than other neural networks. First, the convolutional layer is the core building block. This layer analyzes the height, width, and depth of a 3D image. A feature detector, known as a convolution, will search for specific elements within the image. Filters will receive the image, interpret specific numbers, and correlate them with a feature as the output. With multiple convolutional layers, these CNNs can get more complex, and can produce exact patterns with greater factors. The pooling layer (second) sweeps another filter across the image and essentially reduces the computational complexity and the spatial dimensions. The second layer makes the CNN more efficient and focused on the specific features it is searching for, summarizing the presence of patterns. As hinted by its name, the last output layer essentially connects all the patterns and features identified by the first two layers and gives the final classification of features. During the training process, the CNN is fed a large set of images under an associated label or classification. The system begins learning the classifications of these images, and can identify them quickly after repetitions of training runs. With each training run, the network adjusts its markers, ultimately decreasing error margins and making the technology more accurate. 

IVF is a long process, and one that requires various checkpoints carried out by embryologists. CNNs are beginning to take over the role of embryo morphology decision-making with the ability to carry out visual assessment of scans. Embryo morphology is a step in the IVF process where the features and structure of the embryos are examined to determine if they are suitable to proceed to the next stages of treatment. Using the layers of identification, CNNs can be trained to detect specific features of embryos that may indicate potential health issues or increase the likelihood of failed pregnancy. With extensive training, these AI systems can exponentially reduce human error and variability, ultimately beginning to alter the role of embryologists. AI models, clinically tested at various stages of OVF, are generally creating a more consistent and accurate form of the IVF process. So how do CNNs impact IVF treatments in reality? It all begins at step one.

The Process

Oocytes—more commonly known as egg cells—create the foundation of pregnancies, and thus the IVF process. The success of IVF depends heavily on the quality and condition of the extracted oocytes during the egg retrieval process. Metaphase II oocytes—oocytes that are ready for retrieval and fertilization in the cell process—are critical factors in determining the outcome of embryo development and implantation. Currently, embryologists manually analyze oocytes for quality, but deep learning technologies, more specifically CNNs, are rapidly improving accuracy. This form of AI machine learning is being used to predict likelihood of successful fertilization based on morphological features. CNNs are able to detect the texture, zona pellucida quality (protective layer of an oocyte), and organelle cell structure quality of oocytes. Metaphase II oocytes have specific qualities when ready for fertilization. They must have a smooth exterior layer, a thick zona pellucida to be able to sustain its structure during fertilization, and the correct orientation of chromosomes (genetic material, DNA) in the nucleus. CNNs are making it possible for this screening process to be more comprehensive and precise. 

After the metaphase II oocytes are screened and deemed structurally ready for fertilization, the process and details of fertilization must be secured. Injecting sperm (through an intracytoplasmic sperm injection) is a delicate and complex technique used to directly fertilize individual oocytes. The oocyte must be aligned and stabilized while the egg is punctured and a single sperm is injected. AI can predict, with 99% accuracy, the correct orientation and location for sperm injection. With traditional IVF, embryologists must take time to confidently determine how the injection must be done. CNNs give embryologists the power to quickly determine the proper orientation of the oocytes and location of sperm injection, with increased accuracy compared to human capabilities. 

14-18 hours after insemination (fertilization), a normally fertilized oocyte has 2 pronuclei—the cell structure where DNA is held—within the cytoplasm (from each sex cell – oocyte and sperm). The pronuclei hold the genetic information from both parents and are responsible for creating the single set of chromosomes that the embryo will carry. Abnormal embryos can have 0, 1, or 3 pronuclei. CNNs are applied to identify abnormally fertilized embryos at a 93.1% accuracy rate. After fertilization, embryo cultures are monitored by embryologists for days in laboratories. Only using visual analysis leads to high variability, an issue AI can address. Image classification is being used through machine learning, where direct features are being identified. Recognizing the pronuclei after fertilization indicates if the fertilized oocyte will be ready for embryo development. Being able to detect abnormalities at such an accurate rate allows CNNs to exceed human performance. 

Once the embryo develops, the chromosomes (DNA structures) within the embryo are crucial in ensuring the fetus does not have any genetic abnormalities. Ploidy (# of sets of chromosomes in an organism, humans should have 46) is determined by a screening and genetic testing sequence (PGT-A). The tests are invasive and require a biopsy, which may alter the rates for a successful implantation. AI can conduct the tests noninvasively by analyzing images of the embryo during the IVF process and identify features associated with abnormal ploidy embryos. CNNs can detect features which the human eye cannot, allowing it to analyze images in greater depth. By avoiding the invasive and risky PGT-A test, the process of IVF will become safer and more efficient. 

After the healthy embryos are left for a short amount of time (around 5-6 days), they continue to carry out mitosis (cell division where cells replicate themselves and their genetic material) and grow into blastocysts. A blastocyst is a larger group of cells, with cells beginning to organize themselves into specific groups within the structure based on their location. Each blastocyst is then analyzed to determine which has the highest predicted rate of success in a pregnancy. The chosen blastocyst will then be implanted into the uterus to carry to term. All of the factors combined, as well as blastocyst development prediction, can forecast the development potential of an embryo, and can be used in an algorithm to predict live birth. One AI system, STORK (a deep neural network platform) can combine results from all the previous tests to determine blastocyst quality, with a precision score of 96%, and analyze success rates of implantation. Additionally, the systems AIRE and ERICA are being used to predict positive pregnancy outcomes from embryo morphology using CNNs to analyze blastocyst images with a slightly lower—70-77%—accuracy rate. The AIRE and ERICA assessments give a more accurate prediction of the success rate of pregnancies, creating a greater systematic method for classifications.

In Conclusion

AI and CNN systems are making the IVF process more efficient and accurate. The variability from embryologist to embryologist is eliminated, as this umbrella CNN can identify a multitude of features in the embryo that it can be trained to assess. With each step of IVF, CNNs provide a more accurate and efficient system than individually analyzing scans or carrying out invasive screening procedures. Partnering AI with IVF is creating the most proficient version of infertility treatment, making an incredible discovery even more reliable.