Written by: Courtney Lysiak ‘23
Edited by: Ziwen Zhou ‘23
Medical litigation and poor patient outcomes are serious issues in healthcare. While it’s reasonable to expect varied outcomes to different treatments, obstetrics and gynecology (OB/GYN) is disproportionately affected by these issues. This field is particularly prone to poor patient outcomes due to several challenges, including an incomplete understanding of fetal physiology and the molecular biology of gynecological cancers. Artificial intelligence (AI) may be the key to overcoming many of these issues, leading to better outcomes and fewer malpractice lawsuits.
AI is defined as “the use of complex algorithms in order for machines to reason and perform cognitive functions, including problem solving and decision making,” and involves four elements: machine learning (ML), natural language processing, artificial neural networks (ANNs), and computer vision . Machine learning works by examining patterns and making inferences from wide-ranging data sets and has been observed to outperform other statistical models in its prediction of surgical site infection. It does this by analyzing a variety of data including lab values, diagnosis and treatment . The bulk of applications to obstetrics and gynecology involves technologies in ML and ANNs.
In the field of obstetrics, many birthing injuries can be attributed to misinterpretation of fetal monitoring. Half of hypoxia-induced encephalopathy cases, a brain injury caused by lack of oxygen during birth that leads to lifelong, severe developmental and cognitive impairments, can be attributed to provider mistakes in fetal monitoring . The current prevailing method of fetal monitoring is Cardiotocography (CTG) monitors, which attempt to indirectly assess the development of hypoxia by monitoring fetal heart rate and uterine contractions. However, CTG sensitivity lies at a shockingly low 60% and has not increased since its initial development almost 50 years ago . CTG technology is highly prone to human error and misinterpretation between providers and even when the same provider reviews the results.
In 2010, a group of researchers decided to address some of these issues by taking internatal care in the direction of artificial intelligence through applying principles of supervised machine learning. They converted fetal heart rate variability (HRV) into energy bands of movement, and associated these energy bands with other fetal activity. This information was used to make data segments that would allow the technology to interpret what information was normal and what information was abnormal. Astonishingly, their technology was able to detect pathology half of the time with a low false-positive rate. Combining HRV with other parameters led to more accurate predictions of birth complications . Similar advancements were made in 2014 when researchers compiled a large dataset from CTG monitors to create a random forest classifier (a statistical method of estimation) to identify normal and abnormal CTG patterns with a sensitivity of 72% and specificity of correct CTG interpretation of 78% .
Moving from obstetrics to gynecology, prognosis for gynecological cancer is currently determined using a staged FIGO classification; however, the use of AI may help the field shift towards examining radiological or molecular biomarkers to determine treatment stratifications. One example of this would be the differentiation of pelvic tumors based on extramural vascular invasion. Previous efforts to individualize cancer treatment have faced challenges due to the lack of understanding of the multifaceted mechanisms that are involved in cancer progression. In addition, predictive care in gynecological oncology requires a deep understanding of the molecular mechanisms involved. AI may help alleviate these challenges. Researchers at Imperial College London recently implemented a radiomics-determined mathematical descriptor system to assess epithelial ovarian cancer risk and give better estimations of prognosis . This software was used to evaluate 657 different features pertaining to size, shape, texture, wavelet, and tomographic scans, and was predictive of chemotherapy resistance and poor surgical outcomes.
Artificial neural networks (ANN) have also been applied to IVF treatments to predict embryonic success, defined as delivery at full term. The ANN used four inputs: age of the mother, number of eggs recovered, number of embryos transferred, and whether or not the embryo was frozen, in order to predict outcomes. Time-lapse imaging has also been used to gather large data sets necessary for the success of this technology, which in one groundbreaking study had an 83% accuracy in predicting successful live birth to term by examining 386 images of single blastocyst transfers . Applications of AI in IVF may make the procedure faster, cheaper, and more accessible in the future.
AI may be the face of the future of personalized medicine in obstetrics and gynecology, improving patient outcomes and decreasing litigation. As medicine moves away from outdated models for disease monitoring and intervention, we can expect to see drastic improvements in the care experiences for both clinicians and patients