AI Method for Evaluating Fetal Heart Ultrasound Images
Author Information
Author(s): Aoyama Rina, Komatsu Masaaki, Harada Naoaki, Komatsu Reina, Sakai Akira, Takeda Katsuji, Teraya Naoki, Asada Ken, Kaneko Syuzo, Iwamoto Kazuki, Matsuoka Ryu, Sekizawa Akihiko, Hamamoto Ryuji, Zaffino Paolo, Spadea Maria Francesca
Primary Institution: National Cancer Center Research Institute
Hypothesis
Can an AI-based method improve the assessment of the pulmonary artery-to-ascending aorta ratio in fetal cardiac ultrasound screening?
Conclusion
The AI method using YOLOv7 and UNet3+ significantly improved the accuracy of fetal cardiac ultrasound assessments compared to manual methods.
Supporting Evidence
- The AI method achieved a mean AUC of 0.883, outperforming manual assessments.
- YOLOv7 plus UNet3+ provided better classification of normal and CHD cases compared to other algorithms.
- The study included a diverse dataset of 315 cases, enhancing the reliability of the findings.
- Automated methods reduced variability in measurements compared to manual techniques.
- AI-assisted assessments can support unskilled examiners in clinical settings.
- Standard values for the PA/Ao ratio were established based on the study's findings.
- The study highlights the potential of AI in improving prenatal diagnostic rates for congenital heart disease.
- Limitations included the need for more diverse video data and potential biases in image quality.
Takeaway
This study shows that using AI can help doctors measure the heart's blood vessels in babies before they are born, making it easier to find heart problems.
Methodology
The study involved 315 fetal cardiac ultrasound screenings, using YOLOv7 for image extraction and comparing three segmentation algorithms for accuracy.
Potential Biases
Potential biases include differences in examiner skill levels and variations in video quality due to maternal factors.
Limitations
The study had limited fetal cardiac ultrasound videos, particularly for CHD cases, and the performance may vary based on examiner skill and video quality.
Participant Demographics
The study included 315 women with singleton pregnancies, comprising 280 normal cases and 35 cases of congenital heart disease.
Statistical Information
P-Value
0.883
Confidence Interval
1.237 ± 0.364
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website