Designing and Developing AI-Powered Ultrasound Software for Automatic Fetal Head Biometry Estimation

In routine prenatal ultrasound, it is crucial to assess fetal head biometry for monitoring fetal development, determining gestational age, and detecting potential abnormalities. However, automating the segmentation and measurement of fetal head parameters from ultrasound images remains challenging due to the unique characteristics of these images during different pregnancy trimesters. To address these challenges, we propose an AI-powered ultrasound software. Our solution accurately estimates key parameters such as head circumference (HC), biparietal diameter (BPD), occipitofrontal diameter (OFD), and gestational age using a specialized model based on the encoder-decoder UNet structure. We utilized the training set from the HC18 Grand Challenge dataset for both training and testing our model. To enhance its adaptability to the diverse features present in fetal ultrasound images and to ensure it meets the requirements for a robust deep-learning model, we incorporated augmentation techniques. Our approach achieves excellent segmentation results, with a Dice Similarity Coefficient (DSC) of 97.83 ± 1.28. We further evaluate the model’s performance by calculating the Dice coefficient according to the first, second, and third trimesters and we achieve 96.75 ± 1.47, 97.99 ± 1.27, 97.95 ± 0.75, respectively. After segmentation, we use the least square method for ellipse fitting to measure fetal head circumference. Evaluation metrics for head circumference prediction demonstrate high accuracy, with mean differences and mean absolute differences of 0.16 ± 3.06 and 1.95 ± 2.36, respectively. We then leverage predicted head circumference to estimate gestational age. We have developed a modern Graphical User Interface (GUI) for our model, ensuring flexibility and ease of use. Thorough performance evaluations, including an assessment of inference speed, confirm that our software processes predictions in less than 1 second. In summary, our AI-powered ultrasound software not only overcomes challenges associated with fetal ultrasound image characteristics but also provides a reliable and efficient tool for healthcare providers in monitoring fetal development and making informed clinical decisions.