Artificial Intelligence Improves Placenta Accreta Detection, Perhaps Reducing Maternal Risks
Placenta accreta spectrum (PAS) disorders, where the placenta abnormally implants into the uterine wall, pose important risks to mothers, potentially leading to life-threatening hemorrhage during and after childbirth. These hemorrhages can be uncontrollable, requiring blood transfusions and, in severe cases, even hysterectomy to save the mother’s life.
Currently, diagnosis relies heavily on identifying risk factors like a low-lying placenta (placenta previa) and prior cesarean deliveries, followed by obstetric ultrasound. However, ultrasound interpretation is subjective, dependent on the experience of the sonographer and the quality of the equipment, leading to missed diagnoses until labor begins.
How Artificial Intelligence Learns to Identify Abnormal Placentas
Researchers at Baylor College of Medicine have developed an artificial intelligence (AI) model to improve the detection of placenta accreta. The study,utilizing nearly 40,000 2D obstetric ultrasound images from 113 high-risk pregnant women followed at Texas Children’s Hospital between 2018 and 2025,employed a convolutional neural network – a type of machine learning especially effective at image analysis. This network was trained to differentiate between normal placentas and those exhibiting accreta characteristics.
The AI model doesn’t simply analyze the placenta’s outline. It learns to recognize subtle patterns within the image pixels and integrates clinical data, such as a history of cesarean sections, to calculate a probability score for the presence of placenta accreta.
“Our team is very excited about the potential clinical implications of this model for accurate and rapid diagnosis of placenta accreta.We hope its use as a screening tool will help reduce the maternal morbidity and mortality associated with it,” stated Dr. Alexandra L. Hammerquist, a maternal-fetal medicine specialist at baylor College of Medicine in Houston.
Highly Sensitive Model Awaits Further Validation
Testing of the algorithm revealed an 88% overall accuracy in predicting the presence or absence of placenta accreta. Crucially, the model demonstrated 100% sensitivity – meaning it correctly identified all cases of placenta accreta – and 75% specificity, resulting in only a small number of false positive results and no false negatives.
“The model’s results support its use as a screening tool to reduce missed diagnoses during placenta accreta screening, justifying future prospective trials,” the research team concluded.
Sources:
* Doctissimo: https://www.doctissimo.fr/html/sante/femmes/15013-hysterectomie.htm
* Doctissimo: https://www.doctissimo.fr/html/sante/imagerie/echographie_obstetricale.htm
Worth a look