Classification and Detection of Intracranial Structures

This repository contains a deep learning framework designed for the classification and detection of fetal brain structures in first-trimester ultrasound images. It incorporates a two-stage approach, starting with the classification of ultrasound frames as standard or non-standard using EfficientNet and VGG16, followed by deep feature extraction and LASSO regression for optimal feature selection. The detection stage focuses on multi-label identification of nine key fetal brain structures, using both YOLOv11 for real-time detection and a Swin Transformer model for hierarchical attention-based detection. The framework integrates clinical predictors into logistic regression models, generating nomograms for clinical decision-making.The repository is structured to support modular development, with separate scripts for image preprocessing, classification, feature extraction, LASSO feature selection, nomogram construction, and detection. It also includes performance evaluation metrics for both classification and detection tasks. The complete training pipeline orchestrates the entire workflow, from data loading to model training and visualization generation. The framework is optimized for GPU use and designed to be easily run with Python, requiring specific hardware and software dependencies.