Deep Learning Analysis of Prenatal Ultrasound for Identification of Ventriculomegaly.
Authors
Affiliations (8)
Affiliations (8)
- Department of Systems & Computer Engineering, Carleton University, Ottawa, Ontario, Canada; Department of Methodological & Implementation Research, Ottawa Hospital Research Institute, Ottawa, Ontario, Canada.
- Department of Acute Care Research, Ottawa Hospital Research Institute, Ottawa, Ontario, Canada.
- Department of Acute Care Research, Ottawa Hospital Research Institute, Ottawa, Ontario, Canada; Department of Clinical Science & Translational Medicine, University of Ottawa, Ottawa, Ontario, Canada.
- Children's Hospital of Eastern Ontario Research Institute, Ottawa, Ontario, Canada; Better Outcomes Registry & Network Ontario, Children's Hospital of Eastern, Ottawa, Ontario, Canada.
- Department of Systems & Computer Engineering, Carleton University, Ottawa, Ontario, Canada.
- Department of Systems & Computer Engineering, Carleton University, Ottawa, Ontario, Canada; Department of Methodological & Implementation Research, Ottawa Hospital Research Institute, Ottawa, Ontario, Canada; Children's Hospital of Eastern Ontario Research Institute, Ottawa, Ontario, Canada; School of Epidemiology & Public Health, University of Ottawa, Ottawa, Ontario, Canada; Department of Clinical Science & Translational Medicine, University of Ottawa, Ottawa, Ontario, Canada.
- Department of Acute Care Research, Ottawa Hospital Research Institute, Ottawa, Ontario, Canada; Children's Hospital of Eastern Ontario Research Institute, Ottawa, Ontario, Canada; Better Outcomes Registry & Network Ontario, Children's Hospital of Eastern, Ottawa, Ontario, Canada; Department of Obstetrics & Gynecology, University of Ottawa, Ottawa, Ontario, Canada; School of Epidemiology & Public Health, University of Ottawa, Ottawa, Ontario, Canada; Department of Obstetrics, Gynecology & Newborn Care, The Ottawa Hospital, Ottawa, Ontario, Canada; International & Global Health Office, University of Ottawa, Ottawa, Ontario, Canada.
- Department of Acute Care Research, Ottawa Hospital Research Institute, Ottawa, Ontario, Canada; Department of Obstetrics & Gynecology, University of Ottawa, Ottawa, Ontario, Canada; Department of Obstetrics, Gynecology & Newborn Care, The Ottawa Hospital, Ottawa, Ontario, Canada. Electronic address: [email protected].
Abstract
To develop and evaluate a deep learning model capable of detecting ventriculomegaly on prenatal ultrasound images using a foundation model pre-trained specifically on ultrasound data. A vision transformer-based ultrasound self-supervised foundation model with masked autoencoding (USF-MAE) was fine-tuned for binary classification of fetal brain ultrasound images as normal or ventriculomegaly. The encoder had been pre-trained on more than 370,000 ultrasound images from the OpenUS-46 corpus. For this study, it was adapted and fine-tuned on a curated data set of fetal brain images. Model performance was evaluated through fivefold cross-validation as well as an independent test cohort. Accuracy, precision, recall, specificity, F<sub>1</sub>-score and area under the receiver operating characteristic and precision-recall curves were reported as the performance metrics. Eigen-CAM and Grad-CAM were used to visualize model attention. USF-MAE reported an F<sub>1</sub>-score of 91.76% for the cross-validation set and 91.78% for the test set. The model outperformed all the baseline models, which included VGG-19, ResNet-50, ViT-B/16 and MoCo v3. The model reported a mean test precision of 94.47% and an accuracy of 97.24%. Activation maps indicated that the model consistently focused on the ventricular region when identifying ventriculomegaly. Pre-training on a large, ultrasound-specific corpus improved classification performance and generalization for ventriculomegaly detection. The USF-MAE framework provided strong accuracy, reliability and explainability, showing potential as a robust tool for the assessment of fetal brain structures on prenatal ultrasound.