Hybrid Quantum-Convolutional Neural Network With Spatial Attention for Accurate Classification of Maternal-Fetal Planes in Ultrasound Images

Antenatal ultrasound (US) examination is the primary mode of assessment in fetal development and is considered completely safe for both the mother and the baby. However, ultrasound scans are subject to disturbances by various factors, which complicate the fetal biometry measurements and demand more of the examiner’s time. To enhance the diagnostic efficiency of fetal plane classification for detecting abnormalities and to optimize computational efficiency, a model utilizing quantum computation to accelerate processing speeds in a practical healthcare environment is proposed. A Hybrid Quantum-Convolutional Neural Network with Spatial Attention for Fetal Ultrasound Plane Classification (HQ-CNN-SA) is developed for accurate classification of maternal-fetal planes in ultrasound images is designed for two approaches: (i) classification of pre-processed low-resolution images using HQ-CNN-SA, and (ii) extraction of deep features from high-resolution images by a deep learning model, followed by HQ-CNN-SA classification. It integrates quantum and classical convolution with attention techniques to classify maternal-fetal plane ultrasound images into six categories: maternal cervix, brain, abdomen, femur, thorax, and other. Experimental findings indicate that the low-resolution HQ-CNN-SA model achieves 93% classification accuracy, whereas the high-resolution deep feature-based HQ-CNN-SA scored 95.36% accuracy. Therefore, it effectively handles resolution variability in ultrasound images and improves the reliability of fetal plane classification by supporting more consistent clinical decision-making.