Artificial Intelligence, Intracranial Anomalies, Ultrasound Imaging
Conditions
Brief summary
Timely detection of fetal brain anomalies is critical for improving prenatal counseling and postnatal neurological outcomes. Ultrasonography is the most commonly used and effective imaging method for evaluating fetal structures; however, diagnostic accuracy can be affected by operator experience, fetal position, and image quality, leading to variability in interpretation. Artificial intelligence (AI)-based image analysis offers a new opportunity to standardize diagnostic assessment and reduce subjectivity in ultrasound interpretation. This study aims to evaluate the diagnostic accuracy and clinical applicability of an AI-assisted model (Alyssia) designed to analyze archived 2D fetal brain ultrasound images. The model will be trained and validated to distinguish between normal and abnormal intracranial findings, focusing particularly on the lateral ventricles and other relevant brain regions. The research employs an observational, retrospective design using anonymized ultrasound data obtained during routine prenatal examinations between 18 and 24 weeks of gestation. Expert clinicians will review and label all eligible images to establish ground truth classifications for model training and validation. A deep learning-based algorithm will be developed to automatically classify these images, and its performance will be evaluated using accuracy, sensitivity, specificity, precision, and F1-score metrics. Misclassified cases will be qualitatively analyzed to determine contributing factors such as image quality, anatomical variability, and gestational differences. By comparing AI model outputs with expert-labeled references, the study will assess the model's ability to enhance diagnostic standardization and reduce inter-observer variability. The findings are expected to provide valuable insights into the integration of AI-based decision support systems in prenatal neurosonography. Ultimately, this research aims to support earlier and more reliable detection of fetal brain anomalies, contributing to improved prenatal care and healthier outcomes for mothers and infants.
Interventions
DIAGNOSTIC_TESTAlyssia - AI-Assisted Diagnostic Model for Fetal Brain Ultrasound
Artificial intelligence-based diagnostic tool designed to classify archived 2D fetal brain ultrasound images as normal or abnormal to detect intracranial anomalies.
Sponsors
Sanliurfa Mehmet Akif Inan Education and Research Hospital
Study design
Observational model
COHORT
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
FEMALE
Age
18 Years to 45 Years
Healthy volunteers
Yes
Inclusion criteria
* Archived 2D fetal brain ultrasound images obtained during routine prenatal examinations. * Gestational age between 18 and 24 weeks at the time of imaging. * Maternal age between 18 and 45 years. * Clear visualization of the lateral ventricles and other intracranial regions. * Images meeting diagnostic quality standards suitable for analysis. * Fully anonymized images with no patient identifiers. * Availability of expert assessment to classify each image as normal or abnormal.
Exclusion criteria
* Ultrasound images with poor diagnostic quality or motion artifacts. * Incomplete, duplicate, or corrupted image records. * Ambiguous gestational age or missing clinical metadata. * Images containing any identifiable patient information. * Cases outside the specified gestational window (before 18 or after 24 weeks). * Images unrelated to the fetal brain (misfiled or mislabeled data).
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Diagnostic Accuracy of the AI-Assisted Model (Alyssia) | From study start to model validation (approximately 6 weeks). | The primary outcome is the diagnostic accuracy of the Alyssia artificial intelligence model in classifying archived 2D fetal brain ultrasound images as normal or abnormal. Model performance will be evaluated by comparing AI-generated classifications with expert-labeled ground truth data. |
Countries
Turkey (Türkiye)
Outcome results
None listed