AI Detects Rare Birth Defects in Fetal Ultrasound
AI deep learning identifies rare disorder from first trimester ultrasounds.
Artificial intelligence (AI) deep learning is rapidly emerging as an innovative diagnostic tool for life sciences and health care. A new study demonstrates how AI deep learning can be used to diagnose a rare embryonic developmental disorder called cystic hygroma within the first trimester of pregnancy from fetal ultrasound images.
“In this proof-of-concept study, we demonstrate the potential for deep-learning to support early and reliable identification of cystic hygroma from first-trimester ultrasound scans,” wrote Dr. Mark Walker MD, FRCSC, MSc, MHCM, at the University of Ottawa (uOttawa) Faculty of Medicine and his research team.
Dr. Walker is a perinatologist, a clinical epidemiologist, high-risk obstetrician, co-founder of the OMNI Research Group (Obstetrics, Maternal and Newborn Investigations) at The Ottawa Hospital, which is the largest maternal and newborn research group in Canada, and a professor and the Vice-Dean of Internationalization and Global Health at the uOttawa Faculty of Medicine. He has published over 160 peer-reviewed articles.
Cystic hygroma is caused by malformation or blockage of the lymphatic system that occurs in 1 in 800 pregnancies and 1 in 8,000 live births according to The Fetal Medicine Foundation. A majority of cystic hygroma are located in the neck where there is an excess of fluid accumulation. Cystic hygroma can cause miscarriage or stillbirth.
Around the tenth week of pregnancy prenatal ultrasounds can help diagnose cystic hygroma with the appearance of a space called nuchal thickness, or nuchal translucency (NT).
A sonogram, or fetal ultrasound, is an imaging method that produces images of the fetus in the uterus using sound waves. Fetal ultrasounds are typically conducted the first trimester.
The model was trained using ultrasound image data from The Ottawa Hospital. Specifically, the dataset had 289 sagittal fetal ultrasound images, of which 129 had cystic hygroma, and 160 were normal nuchal thickness controls.
The research team used a Dense Convolutional Network (DenseNet), the DenseNet169 PyTorch model, which connects layers in a feed-forward manner. In artificial intelligence, DenseNet is a type of Convolutional Neural Network (CNN) architecture that reduces the number of parameters, improves feature propagation, address the vanishing-gradient problem, and encourage the reuse of features.
A Convolutional Neural Network is an artificial neural network (ANN) that is often used for computer vision, image, and text classification, as well as object recognition. CNNs contain many layers that learn to spot features of an image.
This study demonstrated that an AI deep learning model could predict cystic hygroma with a 93 percent overall mean accuracy.
Artificial intelligence (AI) deep learning is rapidly emerging as an innovative diagnostic tool for life sciences and health care. A new study demonstrates how AI deep learning can be used to diagnose a rare embryonic developmental disorder called cystic hygroma within the first trimester of pregnancy from fetal ultrasound images.
“In this proof-of-concept study, we demonstrate the potential for deep-learning to support early and reliable identification of cystic hygroma from first-trimester ultrasound scans,” wrote Dr. Mark Walker MD, FRCSC, MSc, MHCM, at the University of Ottawa (uOttawa) Faculty of Medicine and his research team.
Dr. Walker is a perinatologist, a clinical epidemiologist, high-risk obstetrician, co-founder of the OMNI Research Group (Obstetrics, Maternal and Newborn Investigations) at The Ottawa Hospital, which is the largest maternal and newborn research group in Canada, and a professor and the Vice-Dean of Internationalization and Global Health at the uOttawa Faculty of Medicine. He has published over 160 peer reviewed articles.
Cystic hygroma is caused by malformation or blockage of the lymphatic system that occurs in 1 in 800 pregnancies and 1 in 8,000 live births according to The Fetal Medicine Foundation. A majority of cystic hygroma are located in the neck where there is an excess of fluid accumulation. Cystic hygroma can cause miscarriage or stillbirth.
Around the tenth week of pregnancy prenatal ultrasounds can help diagnose cystic hygroma with the appearance of a space called nuchal thickness, or nuchal translucency (NT).
A sonogram, or fetal ultrasound, is an imaging method that produces images of the fetus in the uterus using sound waves. Fetal ultrasounds are typically conducted the first trimester.
The model was trained using ultrasound image data from The Ottawa Hospital. Specifically, the dataset had 289 sagittal fetal ultrasound images, of which 129 had cystic hygroma, and 160 were normal nuchal thickness controls.
The research team used a Dense Convolutional Network (DenseNet), the DenseNet169 PyTorch model, which connects layers in a feed-forward manner. In artificial intelligence, DenseNet is a type of Convolutional Neural Network (CNN) architecture that reduces the number of parameters, improves feature propagation, addresses the vanishing-gradient problem, and encourages the reuse of features.
A Convolutional Neural Network is an artificial neural network (ANN) that is often used for computer vision, image, and text classification, as well as object recognition. CNNs contain many layers that learn to spot features of an image.
This study demonstrated that an AI deep learning model could predict cystic hygroma with a 93 percent overall mean accuracy.
“Our findings demonstrate the feasibility of using deep-learning models to interpret fetal ultrasound images and identify cystic hygroma diagnoses with high performance in a dataset of first-trimester ultrasound scans,” the researchers concluded.
This scientific study is just the start. According to the researchers, with additional testing with a large multi-site dataset and external validation, their AI deep learning model can be used to predict other fetal anomalies from ultrasonography.
Source: https://www.psychologytoday.com/us/blog/the-future-brain/202209/ai-detects-rare-birth-defects-in-fetal-ultrasound