Related Experiment Video
Updated: May 21, 2025
08:04
Murine Fetal Echocardiography
Published on: February 15, 2013
17.2K
Fetal Diagnostics using Vision Transformer for Enhanced Health and Severity Prediction in Ultrasound Imaging
Eshika Jain1, Pratham Kaushik1, Vinay Kukreja1
1Center for Research Impact & Outcome, Chitkara University Institute of Engineering and Technology, Chitkara University, Punjab, India.
Current Medical Imaging
|March 19, 2025
Summary
This study introduces a new Vision Transformer (ViT) system for fetal ultrasound analysis, achieving 90% accuracy in classifying fetal health and abnormality severity. This AI-driven approach promises to enhance early detection and reduce neonatal mortality rates.
Area of Science:
- Medical Imaging and Artificial Intelligence
- Fetal Health Monitoring
- Deep Learning in Healthcare
Background:
- Neonatal mortality remains a challenge, particularly in developing nations, necessitating advanced diagnostic tools.
- Existing machine learning models for fetal ultrasound require enhanced precision for accurate prediction.
- Artificial intelligence offers a promising avenue for improving the accuracy of fetal health assessments.
Purpose of the Study:
- To develop and evaluate a novel health classification and severity detection system using Vision Transformers (ViTs) for fetal ultrasound imagery.
- To improve the precision of fetal health status detection and abnormality assessment compared to traditional models.
- To explore the efficacy of ViTs in discerning intricate patterns within fetal ultrasonographic imagery for precise categorization of fetal well-being.
Main Methods:
- A dataset of 500 fetal ultrasound images was collected and annotated by radiologists for health status and abnormality severity.
- The dataset underwent various pre-processing techniques before model training.
- An optimized Vision Transformer (ViT) model was trained using the Adam algorithm with a Cross-Entropy loss function.
Main Results:
- The developed ViT model achieved 90% classification accuracy for detecting severity.
- The model demonstrated a high F1-score of 0.87 and a Mean Absolute Error (MAE) of 0.30.
- ViTs showed superior efficacy in capturing fine-grained spatial relationships in ultrasound images, leading to accurate predictions.
Conclusions:
- Vision Transformers (ViTs) show significant potential to revolutionize fetal health monitoring through accurate and reliable AI-driven predictions.
- The system can contribute to reducing neonatal mortality by enabling early interventions based on precise diagnostic information.
- This research establishes a benchmark for future AI applications in fetal health diagnostics.
