The Evolution of Gynecologic Ultrasound and AI Integration
Ilan Timor-Tritsch, MD, FAIUM
Hackensack Meridian School of Medicine, New Jersey
Department of Ob/Gyn, Mount Sinai Medical Center, New York
Maternal Resources, New Jersey
Dr. Ilan Timor-Tritsch is a distinguished figure in obstetrics and gynecology, having graduated from the Hebrew University in Jerusalem in 1962. After serving in the Israeli military, he completed his OBGYN residency in Haifa, Israel. His interest in electronics and engineering led him to spend a year conducting research at the Israeli Technion. Through a connection with an engineer from his naval service, he helped develop one of the first transvaginal ultrasound probes in Israel during the late 1970s. Dr. Timor is widely recognized for introducing transvaginal ultrasound in the United States, having authored over 150 articles on the subject. His early work with pattern recognition in ultrasound laid the groundwork for today’s AI applications in the field.
How has traditional ultrasound practice evolved in gynecology over the years?
The history of gynecological ultrasound over the past 40 years has seen significant changes. Initially, OBGYNs lost ground to radiologists because they weren’t interested in the technology. While this has largely changed in Europe and many other countries, gynecological ultrasound in the United States is still predominantly performed by radiologists. I’ve written several articles voicing my disagreement with this practice. I believe gynecological ultrasound is more than just imaging – it’s an extension of the bimanual pelvic exam. The transvaginal approach complements the pelvic examination by providing detailed visualization of the ovaries and other pelvic structures. Even at major institutions where I’ve worked, like Columbia Presbyterian, NYU, and Hackensack Medical Center, most gynecologic ultrasound is still outsourced to radiology. At NYU, I managed to reclaim about 60% of these procedures during my 21-year tenure.
What are some key advancements in AI and automation that are transforming gynecologic ultrasound?
AI offers numerous advantages in diagnosis, treatment planning, and patient care. It improves interpretation through pattern recognition, particularly in detecting and characterizing gynecologic pathology, especially ovarian lesions and endometriosis. One specific example is the detection of borderline ovarian tumors, which show typical microcystic appearances in their solid components. These tumors primarily affect young women, and early detection through AI-assisted recognition could help primary care gynecologists make timelier diagnoses. AI can also automate measurements of structures like the uterus, cervix, ovarian volume, and endometrial thickness, making these assessments faster and more reliable than manual measurements.
The technology can generate automatic reports and potentially predict outcomes, particularly in differentiating between benign and malignant ovarian tumors. This ties into the work of the International Ovarian Tumor Analysis (IOTA) group, led by Dirk Timmerman in Europe, which has established clear criteria for distinguishing benign from malignant features that AI can recognize. Furthermore, AI can provide real-time feedback during examinations, potentially shortening the time between examination and diagnosis. It can suggest additional views or measurements that might be needed, improving the completeness of the examination.
How do you see AI algorithms improving the accuracy and efficiency of ultrasound imaging?
AI can provide more precise and accurate management by delineating the borders of different structures through analysis. I’ve observed this capability in some current machines that can evaluate the uterus, perform measurements, and assess tissue structure almost instantaneously. This provides immediate feedback that enhances diagnostic accuracy. A particularly interesting application is in follicular counting for fertility treatment. While early attempts at automated 3D follicle counting had limitations due to resolution issues, today’s high-resolution transvaginal ultrasound, combined with AI, can better differentiate and measure individual follicles. In the past, when follicles were close together, they would appear merged in the image, making accurate counting impossible. However, with today’s improved resolution and AI assistance, we can clearly distinguish separate follicles, measure them, and track them over time. This is especially valuable for reproductive endocrinologists monitoring patients undergoing fertility treatments.
What do you see as the major barriers to widespread adoption of AI in ultrasound imaging?
First, practitioners need to be convinced that AI is reliable and based on actual clinical data, especially given recent controversies about AI in other fields. There’s considerable skepticism about AI’s accuracy and reliability, partly due to its misuse in other contexts. We need to demonstrate that medical AI applications are based on sound, real-world clinical data. Second, users need proper training in how to utilize AI tools effectively. When I recently examined a Samsung system with AI capabilities for uterine measurements, it became clear that practitioners need to know how to engage with the system and ask the right questions. Application specialists who provide training when new ultrasound systems are installed are crucial in this process. They serve as essential intermediaries between manufacturers and clinicians, teaching us not only what AI can do but also when and how to use it effectively. Clinicians need to understand what algorithms are available, which cases they apply to, how results are displayed, and how reliable the information is. For example, if AI reports that an ovary has 15 follicles greater than 10-millimeters, clinicians need to know how reliable that count is. Without this knowledge transfer, many valuable AI capabilities may go unused.
Can you share any personal experiences with using AI in your research?
I recently wrote an article about cesarean scar pregnancy, a controversial topic where there’s debate about whether to classify it as an ectopic pregnancy. I used AI to help me analyze the different perspectives and find ways to reconcile opposing viewpoints. The AI provided fact-based, research-supported insights that I hadn’t considered, though I had to modify and adapt these insights rather than use them directly. While journal editors are increasingly requiring authors to disclose AI use, I found it valuable as a tool for expanding my thinking and generating new approaches to complex clinical issues.