As technology advances, the healthcare field continues to find innovative ways to improve patient outcomes and enhance medical procedures. One area of medicine that has seen significant advancements in technology is obstetrics and gynecology. With the integration of artificial intelligence (AI) into fetal ultrasound procedures, healthcare providers are now able to access a higher level of accuracy and efficiency in identifying potential issues. In this article, we will explore the benefits of AI in fetal ultrasound.
According to an article published in Ultrasound in Obstetrics & Gynecology in 2020, while obstetric and gynecological ultrasound are among the most commonly performed imaging studies, AI has yet to make significant inroads in this field.
Artificial intelligence has shown promise in transforming the healthcare sector by uncovering key insights from the vast amount of digital data generated during healthcare delivery. Most of these efforts are focused on screening, prediction, triage, diagnosis, drug development, treatment, monitoring, and imaging interpretation.
Nonetheless, there are some studies focusing on obstetrics and gynecology that have utilized AI to evaluate adnexal masses, assess aneuploidy risk, predict fetal lung maturity, estimate gestational age, and classify fetal brain images as normal or abnormal.
The Challenges of AI in Obstetrics and Gynecology
Ultrasound is one of the most used imaging methods for obstetrics and gynecology, and in the article Introduction to Artificial Intelligence in Ultrasound Imaging in Obstetrics and Gynecology, the authors explained that in comparison to analyzing a CT scan, ultrasound AI software must fit into the workflow differently due to the need for real-time analysis at the point of acquisition.
The researchers stated that recent advancements in obstetric and gynecological ultrasound have shown promising potential for automated detection of standard planes, quality and assurance in fetal ultrasound.
Moreover, automation in ultrasound technology can reduce the scan duration by eliminating repetitive tasks such as acquiring standard planes and adjusting calipers, freeing up more time for additional scan planes or communication with patients. This is particularly important given the global shortage of imaging experts and the growing demand for diagnostic imaging.
While artificial intelligence has the potential to aid in clinical diagnosis and management, there are concerns regarding its applicability in certain medical contexts. Imaging features alone may not be sufficient to determine a diagnosis or proper treatment plan.
This highlights the importance of understanding when AI can and cannot be applied in medical settings and the need for research in building AI models that integrate imaging and electronic health record data for personalized diagnostic imaging.
AI Assistance in Fetal Ultrasound During the First Trimester
Gestational Sac
In an article published in Frontiers in Medicine in 2021, the authors explained that the gestational sac (GS) is an important structure observed during ultrasound in pregnancy, and its diameter can roughly estimate the gestational age. There’s an automatic solution to select the standardized biometric plane of the GS and perform measurements during routine ultrasound examinations, which showed robustness, efficiency, and accuracy in both quantitative and qualitative analysis.
While this algorithm is limited to normal gestation within 7 weeks, it has potential to facilitate clinical workflow with further validation. Additionally, the developers established a fully automated framework for simultaneous semantic segmentation of multiple anatomical structures including the fetus, gestational sac, and placenta, demonstrating superior segmentation results, good agreement with expert measurements, and high consistency against scanning variations.
Fetal Biometry Assessment
The authors of Artificial Intelligence in Prenatal Ultrasound Diagnosis considered that automating image-based assessments of fetal anatomies in the initial trimester is a challenging and understudied area. This led to the development of an intelligent image analysis method to automate biometry and visualize key fetal anatomy in the first trimester, requiring only a simple standard acquisition guideline for a 3D ultrasound scan.
The method performed semantic segmentation of the whole fetus and extracted biometric planes of the head, abdomen, and limbs for anatomical assessment. However, the study found relatively low qualitative analysis results for the limbs due to a low detection rate.
Nuchal Translucency
The nuchal translucency (NT) is a fluid-filled area under the skin of the posterior neck of a fetus that plays a crucial role in prenatal screening for chromosomal abnormalities, congenital heart disease, and intrauterine fetal death. In the article mentioned above, the authors explained that measuring the NT thickness accurately is challenging due to factors such as low signal-to-noise ratio, short fetal crown-rump length, and activity in early pregnancy.
Initially, manual and semi-automatic approaches were used to measure NT thickness, but recent breakthroughs in automation have enabled accurate automatic identification and measurement of the standard NT plane thickness using mid-sagittal section images and automatic recognition methods.
AI Assistance in Ultrasound During the Second and Third Trimester
Biometric Measurement
Prenatal ultrasound relies on standardized measurements to date pregnancies and detect potential abnormalities, but this process can be highly repetitive. Researchers of the article published in Frontiers in Medicine proposed automation as a solution that can reduce the time needed for these routine tasks, freeing up more time for analysis of additional scan planes for diagnosis. In addition, automated measurements can reduce operator bias and improve quality control.
Fetal Head Measurement
To measure fetal head circumference, there are several AI-based methods that have been developed. The authors of Artificial Intelligence in Prenatal Ultrasound Diagnosis explained that these methods consider various factors that may affect detection and measurement accuracy, such as abnormalities, low contrast, speckle noise, boundary occlusion, or artifacts.
Some approaches use plane verification for more accurate results, such as assessing the transthalamic plane on the basis of the cavum septum pellucidum, the V-shaped ambient cistern, and the cerebellum. Other methods combine intelligent processing with the Hadlock curve to determine gestational age automatically. Medical devices equipped with software for intelligent processing have also proven reliable in measuring head circumference, showing potential for improving workflow efficiency with further optimization.
Fetal Abdominal Circumference
Measuring the abdominal circumference (AC) in fetal ultrasound is challenging due to low contrast, irregular shape, and high variability of images. Researchers in the article mentioned above stated that several studies have been conducted on this topic, leading to the development of a novel method using the spine position as a navigation marker to determine the final plane for AC measurement, reducing interference from factors such as lack of amniotic fluid or acoustic shadows.
The standard plane checking process improves stability and accuracy of abdominal circumference measurements. This intelligent approach significantly outperforms conventional studies and shows potential for integration into a single framework using a multiple learning framework.
Fetal Long Bone
Fetal long bone measurement is also challenging due to variations in fetal position and posture, according to the same authors. There have been advancements in segmenting and measuring the fetal femur, with methods involving determination of regions of interest, image processing, identification of femoral features, and measurement of lengths or volumes. These models have similar accuracy to manual measurements.
A prospective study included in the article published in Frontiers in Medicine evaluated the performance of a 3D ultrasound system, five-dimensional long bone (5DLB), in detecting lower limb long bone and found it to be reproducible and comparable to conventional 2D and manual 3D techniques for fetal long bone measurements. The new technique streamlines the process of reconstructing images and performing fetal biometry.
Prenatal Ultrasound Diagnosis with AI Assistance
Diagnosing Respiratory Diseases
In an article published in the Journal of Clinical Medicine in 2023, the authors explained that assessing fetal lung maturity is crucial in predicting premature mortality and neonatal respiratory morbidity, but current methods such as amniocentesis are invasive and may yield inaccurate results. Conventional ultrasound can predict fetal lung maturity noninvasively, but limitations such as subjectivity and maternal-fetal status hinder its clinical application.
Texture feature analysis can reduce subjective examiner variation and effectively quantify fetal lung maturity. An automatic quantitative ultrasound analysis (AQUA) texture extractor for fetal lung maturity quantification with high accuracy was proposed in the article, along with other studies that have developed AI-based models such as quantusFLM to predict the occurrence of respiratory distress syndrome in newborns with comparable accuracy to amniotic fluid tests. These models showed potential for immediate clinical application and good stability and reproducibility.
Diagnosing Intracranial Malformations
Central nervous system (CNS) malformations are common and can be diagnosed by non-invasive methods such as fetal neurosonography (NSG), but incorrect fetal head position, maternal obesity, and lack of expertise can affect the quality and accuracy of imaging.
The authors of Application and Progress of Artificial Intelligence in Fetal Ultrasound believe that AI-assisted ultrasound diagnosis can help overcome these limitations and improve detection rates. They explained that some studies have developed algorithms and models using deep learning networks such as U-Net and VGG-Net to distinguish normal from abnormal fetal brain images.
These models can reduce false-negative rates and visualize lesion sites through heat maps and overlapping images. Researchers also added that other studies have developed AI-assisted image recognition systems such as PAICS, which can detect and classify nine kinds of fetal brain malformations in real-time with comparable accuracy to experts. With its significant progress in this field, AI is expected to become an effective tool for clinically screening fetal CNS malformations.
Diagnosing Congenital Heart Diseases
In the last article mentioned, the authors stated that congenital heart disease (CHD) is a severe and common congenital disease among newborns, which can lead to costly surgical treatments, long cycles of treatment, the risk of secondary surgery, and high mortality.
Prenatal ultrasound diagnosis of fetal CHD can assist in making clinical decisions and improve neonatal outcomes, but identifying complex abnormal fetal heart anatomy is challenging. However, the authors note that in recent years, AI techniques have made significant progress in assessing cardiac structure and function, proving to be a promising tool for diagnosing congenital heart disease in fetuses.
The authors also added that several studies have validated the potential of AI to diagnose fetal CHD, shortening training periods, reducing subjectivity, and improving diagnostic accuracy. However, challenges remain due to factors that impact fetal ultrasound image quality.
Artificial intelligence has made significant strides in the field of fetal ultrasound to help practitioners in diagnosing conditions and also observe the normal development of the pregnancy. However, further exploration is needed to address the challenges mentioned at the beginning of this article.
It is important to highlight that another benefit from intelligent algorithms includes the improvement of workflow and the reduction interruptions during live scanning. The authors of an article published in Prenatal Diagnosis in 2021 believe the reduction in task switching and interruption of observational scanning can transform the sonographic experience by allowing sonographers to automate repetitive thereby to focusing on other aspects of the scan and enhancing scanning quality and prenatal care for families.
Still, large-scale trials are needed to determine if AI assistance increases antenatal fetal anomaly detection rates.