Early Detection of Fetal Brain Abnormalities using CNN Framework

This research introduces an innovative approach to the early detection and classification of fetal brain abnormalities through the integration of ResNet-50 Convolutional Neural Network (CNN) framework. Central to our methodology is the strategic incorporation of Gray Level Co-occurrence Matrix (GLCM) feature extraction, enhancing both the accuracy and time efficiency of automated prenatal diagnostics. Leveraging the deep learning capabilities of ResNet-50, our model adeptly captures intricate hierarchical features essential for precise classification, forming a robust foundation for nuanced analysis of fetal brain MRI scans. Empirical validation on a diverse dataset underscores the model's superior performance, exhibiting heightened accuracy in early detection and classification compared to existing methodologies. Particularly noteworthy is the role of GLCM in optimizing time efficiency without compromising diagnostic precision, a critical practical consideration in healthcare decision-making. This research stands as a significant contribution to the field of automated prenatal diagnostics, offering a powerful tool that seamlessly integrates advanced deep learning techniques with efficient texture analysis, promising transformative impacts on timely and informed healthcare decision-making, the classification model performance is evaluated using precision, recall, and F1-score metrics for five different classes representing medical conditions. Class 0, representing normal cases, exhibits a precision of 0.75, indicating that 75% of predicted normal instances are accurate, while a recall of 0.83 suggests that 83% of actual normal cases are correctly identified. Class 1 (septi pellucidi) demonstrates a precision of 0.80, indicating 80% accuracy in predictions, and a recall of 0.73, signifying that 73% of true positive instances are captured. Class 2 (colpocephaly) boasts a high precision of 0.89, denoting 89% accuracy in positive predictions, and a recall of 0.85, revealing an 85% true positive capture rate. Class 3 (hypoplasia) and Class 4 (polymicrogyria) both exhibit precision and recall values above 0.85, suggesting strong model performance in identifying these conditions. The macro and weighted average F1-scores are both 0.82, indicating an overall balanced performance across all classes. The model achieves an accuracy of 82%, highlighting its ability to correctly classify instances across the diverse set of conditions.