Improved Prenatal Brain Anomaly Detection Using Xception Architecture with Attention Mechanisms and Region-Based Segmentation

Neo-natal brain problems complicate prenatal care, which requires prompt and precise diagnosis to ensure proper treatment. Deep learning-based techniques present a viable remedy by automating the identification of prenatal brain anomalies. A novel framework combining the Xception architecture with attention processes improves anomaly detection by giving priority to diagnostically important regions.While depth-wise separable convolutions enhance feature extraction, spatial attention approaches enhance accuracy and interpretability significantly. Important anatomical features are separated using region-based segmentation in this methodology. The AdamW optimizer ensures improved generalization and training efficacy. In experiments, the accuracy rate 92.5% and computational efficiency of identifying neural tube defects and ventriculomegaly become significantly higher than with traditional convolutional neural networks.