Artificial Intelligence to Determine Fetal Sex.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Obstetrics and Gynecology

Background:

• Fetal sex determination is crucial for reproductive health and genetic counseling.
• Current methods rely on skilled ultrasound operators, limiting accessibility in some regions.
• Advancements in artificial intelligence (AI) offer potential for automated analysis of medical images.

Purpose of the Study:

• To evaluate the performance of a proof-of-concept AI model in accurately determining fetal sex from ultrasound images.
• To assess the confidence and accuracy of AI-driven fetal sex identification.
• To explore the potential of AI integration in routine obstetric care.

Main Methods:

• A computer vision model utilizing EfficientNetB4 architecture was trained on 11,769 ultrasound image slices.
• The model was tested on a separate dataset of 7,443 image slices from a high-volume practice.
• Performance was evaluated using Accuracy, Cohen's Kappa, and Receiver Operating Characteristic Area Under the Curve (ROC AUC).

Main Results:

• The AI model achieved an Accuracy of 88.27% and a Cohen's Kappa score of 0.843 on the test set.
• ROC AUC scores for Male, Female, and Unable to Assess categories were 0.896, 0.897, and 0.916, respectively.
• A high ROC AUC score of 0.981 was achieved for the Text Added category.

Conclusions:

• This novel AI model demonstrates a high capability for fetal sex determination from ultrasound data.
• The findings support the integration of AI into obstetric workflows, enhancing diagnostic capabilities.
• This technology could significantly benefit areas with limited access to specialized ultrasound expertise.