Dynamic graph-based quantum feature selection for accurate fetal plane classification in ultrasound imaging

Area of Science:

• Medical Imaging
• Quantum Computing
• Artificial Intelligence

Background:

• Accurate classification of fetal biometric planes is crucial for prenatal screening and diagnosing fetal abnormalities.
• Current methods may lack efficiency and interpretability in analyzing complex ultrasound data.

Purpose of the Study:

• To propose a novel Dynamic Graph-Based Quantum Feature Selection (DG-QFS) framework to enhance fetal ultrasound image classification.
• To improve diagnostic efficiency and accuracy in prenatal screening.

Main Methods:

• Features were extracted from ultrasound images using a pre-trained deep learning model.
• A quantum-driven feature selection pipeline modeled inter-feature relationships using dynamically entangled multi-qubit graphs.
• Qubits represented feature nodes, with entanglement scores and graph centrality guiding feature selection.

Main Results:

• The DG-QFS method achieved a classification accuracy of 96.73% on a dataset of 12,400 fetal ultrasound images.
• The model demonstrated superior performance compared to baseline deep learning and conventional feature selection techniques.
• Improvements were noted in accuracy, generalization, and interpretability.

Conclusions:

• The proposed DG-QFS framework effectively enhances fetal ultrasound image classification accuracy.
• Integrating quantum computing principles offers a promising avenue for improving prenatal diagnostic tools.
• The method shows potential for more accurate and interpretable early detection of fetal abnormalities.