OC07.07: Developing a transfer learning model for the prediction of Down syndrome from first trimester ultrasound images

This study aims to develop a transfer learning model for automated screening of Down syndrome using first trimester ultrasound images. An artificial intelligence-assisted diagnostic ultrasound study was conducted between January 2019 and March 2021. Ultrasonographic images were collected from pregnant women undergoing aneuploidy screening during the first trimester of pregnancy. We utilised a transfer learning model based on the AlexNet architecture, implemented in MATLAB software, for model training and validation. The class activation map (CAM) was employed to detect the key features contributing to the diagnostic process of the model. The confusion matrix and the receiver operating characteristic (ROC) curve, along with the area under the ROC curve (AUC), were utilised to evaluate the model performance. The dataset comprised 1056 images of normal fetuses and 12 images of fetuses with Down syndrome. To address class imbalance, data augmentation techniques were employed, expanding the 12 Down syndrome images to 96. The CAM visually emphasised critical regions within the input ultrasound image in nuchal translucency and the nasal bone. The model demonstrated exceptional performance, achieving an overall accuracy of 98.26%. The AUC result was 0.98, accompanied by high sensitivity (89.47%), specificity (100%), positive predictive value (100%), and negative predictive value (97.96%). The transfer learning model proves valuable for early detection of Down syndrome. This approach holds promise in reducing time consumption and errors in manual measurements while maintaining a high level of accuracy. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.