TL;DR:
- Artificial intelligence (AI) in obstetrics is rapidly evolving, offering advancements in ultrasound imaging, fetal heart rate interpretation, and predictive modeling.
- AI in obstetric ultrasound enables probe guidance, biometric plane identification, anomaly detection, and real-time education for sonographers and trainees.
- Fetal heart rate interpretation using AI involves automating alerts and identifying novel patterns associated with fetal distress.
- Predictive modeling with AI aids in assessing the risk of adverse perinatal outcomes, including preeclampsia, shoulder dystocia, and postpartum depression.
- Natural language processing, a form of AI, shows promise in analyzing clinician notes and social media data for maternal morbidity prediction and pregnancy drug safety.
- While AI presents opportunities for improving diagnosis, patient care, and workflow efficiency, concerns about fairness, accountability, transparency, and ethics must be addressed.
- Clinicians need to critically evaluate the development and implementation of AI models, considering population similarities and meaningful prediction targets.
- AI’s impact on the market necessitates cautious and responsible adoption, with healthcare organizations and policymakers striving for fairness and equity in AI applications.
Main AI News:
The introduction of ChatGPT, a groundbreaking generative artificial intelligence (AI) system, has disrupted the field of medical education since its launch in November 2022. ChatGPT has displayed remarkable capabilities, including drafting insurance referral letters, passing medical licensing examinations, and generating convincing chest X-ray images. Moreover, this AI marvel has even fooled scientific journal editors with its production of authentic-looking scientific abstracts. The impact of AI is already evident in our daily lives through smart home devices, real-time navigation software, and online shopping recommendations. Yet, its integration into clinical medicine has quietly progressed, necessitating a closer examination of how AI is transforming obstetrics and the potential implications for healthcare practitioners.
AI, defined as the development of computer systems capable of human-like tasks, encompasses machine learning—a process in which computers learn independently, devoid of predetermined rules. Machine learning is further divided into three subcategories: supervised learning, unsupervised learning, and reinforcement learning. Neural networks underpin these subdivisions, simulating the functions of neurons by analyzing input variables, activating functions, and generating predictions that refine accuracy over subsequent cycles.
AI has already established a strong presence in hospital operations and insurance risk prediction, revolutionizing the field of clinical medicine. In pathology and laboratory medicine, AI-driven automated systems play a crucial role in the initial triage of Papanicolaou test cytology abnormalities. Similarly, AI enhances radiology workflows by facilitating study triage based on acuity and optimizing study protocoling and radiation control. These examples merely scratch the surface, as the FDA has approved numerous clinical AI algorithms and applications for medical care. However, in the realm of obstetrics, the incorporation of AI remains in its nascent stages. Nonetheless, AI adoption in obstetrics has shown rapid growth, as evidenced by its utilization in obstetric ultrasound, fetal heart rate interpretation, and the prediction of adverse clinical outcomes.
In obstetric ultrasound, AI offers various potential applications, many of which have already become a reality. These include probe guidance, fetal biometric plane identification, anomaly scan completeness, anomaly highlighting, and the enhancement of educational opportunities for sonographers and trainees. Recent advancements demonstrate AI’s proficiency in identifying biometric planes, detecting anatomic malformations, completing comprehensive anatomy ultrasounds, and even classifying full fetal echocardiograms. Notably, a neural network trained to estimate gestational age has exhibited superior accuracy compared to trained sonographers conducting standard fetal biometry. As innovation and interest from clinicians continue to drive progress, the integration of AI into obstetric ultrasound is poised for continued expansion.
The application of AI in fetal heart rate (FHR) interpretation has been a topic of exploration since the advent of continuous electronic fetal monitoring. AI’s potential in this area can be categorized into two primary objectives: automating alerts based on predefined classification systems and identifying novel patterns indicative of fetal distress. While automating alerts has shown promise in predicting fetal distress, larger clinical trials have yielded less impressive results.
However, the evaluation of FHR tracings for abnormal patterns has displayed greater potential. Research efforts have produced systems capable of identifying novel findings, such as decelerative capacity, which involves assessing the frequency, depth, and slope of decelerations in the fetal heart rate. Although these systems are not yet ready for widespread adoption, their ability to uncover previously unnoticed patterns necessitates further investigation.
Predictive modeling has long been employed by obstetricians to identify individuals at risk for adverse perinatal outcomes. Recent years have witnessed the rise of AI-driven predictive models, which differ from traditional models in that they allow computers to learn clinical patterns and make predictions independently. These AI models have demonstrated their efficacy in predicting conditions ranging from preeclampsia and shoulder dystocia to postpartum hemorrhage and depression, as well as determining the optimal modes of conception and delivery.
Another emerging application is natural language processing, which enables the interpretation of text in a manner akin to human understanding. Natural language processing has shown promise in predicting severe maternal morbidity using clinician notes and mining social media data for pregnancy drug safety. Its potential extends far beyond these initial explorations and warrants further exploration.
While AI presents tremendous opportunities to enhance disease diagnosis, patient care, and clinician workflows in obstetrics, it also raises legitimate concerns. Cathy O’Neil, MD, in her book “Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy,” highlights the dangers of opaque, scalable, and feedback-loop-enabled AI, characteristics that can be applied to clinical medicine. For instance, the well-known vaginal birth after cesarean (VBAC) calculator inadvertently perpetuated racial disparities by incorporating recommendations biased against Black and Hispanic patients.
To mitigate such risks and ensure fairness, accountability, transparency, and ethical principles in AI, various organizations have adopted guidelines. Additionally, the US federal government aims to establish formalized guidance through the Agency for Healthcare Research and Quality, addressing the impact of healthcare algorithms on racial and ethnic disparities. Until these measures are formalized, healthcare practitioners must remain vigilant and comprehend the potential consequences of the indiscriminate application of algorithms, regardless of whether they are generated using traditional models or AI-driven techniques.
It is imperative for obstetricians to critically evaluate novel technologies, including AI, within their models of care. When considering predictive modeling, clinicians should scrutinize the development process and assess the risk of bias, regardless of whether traditional or AI-driven approaches were employed. Key considerations include the similarity between the model’s development population and the clinician’s patient population, as well as the meaningfulness of the prediction targets. Some models rely solely on the International Classification of Diseases, Tenth Revision coding, which may lead to false negatives and positives. Prior to the widespread implementation of predictive modeling or AI on a large scale, clinicians must engage actively and gain a thorough understanding of these critical factors.
Conclusion:
The integration of artificial intelligence into obstetrics represents a transformative development with vast implications for the market. The rapid progress in AI-driven applications, such as ultrasound imaging, fetal heart rate interpretation, and predictive modeling, offers unprecedented opportunities for improving clinical decision-making and patient outcomes. However, market players need to navigate potential challenges related to fairness, accountability, transparency, and ethical considerations. By embracing AI responsibly and ensuring that guidelines promote equity, healthcare organizations can leverage this technology to drive innovation, enhance patient care, and stay at the forefront of the evolving market landscape.
