Technical implementation of the AI-based ICH detection system
The RSNA Intracranial Hemorrhage dataset is a large-scale, expertly annotated collection of brain CT scans specifically curated for developing and evaluating automated ICH detection algorithms. Each scan has been reviewed and labeled by board-certified radiologists, ensuring high-quality ground truth annotations.
Medical CT scans require specialized preprocessing to enhance hemorrhage visibility and prepare data for deep learning models.
System accepts DICOM (.dcm) files, the standard format for medical imaging. DICOM contains both image data and metadata (patient information, scan parameters).
Raw pixel values are converted to Hounsfield Units (HU), a standardized scale for measuring radiodensity. This normalization ensures consistency across different scanners and imaging protocols.
HU = pixel_value × RescaleSlope + RescaleIntercept
Three different HU windowing settings are applied to emphasize different tissue types, creating a pseudo-RGB representation:
WL: 40 HU | WW: 80 HU
Optimizes visualization of brain parenchyma and gray-white matter differentiation
WL: 75 HU | WW: 215 HU
Enhances detection of acute hemorrhage and blood products
WL: 600 HU | WW: 2800 HU
Shows skull structures and helps identify fractures
The three windowed images are stacked to create a 3-channel RGB-like tensor:
This multi-window approach provides the neural network with comprehensive information about different tissue densities simultaneously, improving hemorrhage detection across all subtypes.
Images are normalized to [0, 1] range and resized to 256×256 pixels for efficient neural network processing while maintaining diagnostic features.
Different ICH subtypes have varying radiodensities and locations. By using three complementary windows, the system can detect hemorrhages regardless of their density or proximity to bone, significantly improving sensitivity and specificity compared to single-window approaches.
Hybrid Deep Learning Architecture for Multi-Label ICH Classification
The system employs a novel cascade architecture that combines the strengths of two state-of-the-art convolutional neural networks in a sequential refinement process.
Role: Initial feature extraction and local pattern recognition
EfficientNet-V2 provides superior parameter efficiency while maintaining high accuracy. Its compound scaling approach ensures optimal resource utilization, critical for medical imaging applications requiring detailed feature extraction.
Role: Global spatial context understanding and feature refinement
ConvNeXt modernizes standard CNNs with Vision Transformer-inspired improvements while maintaining the efficiency and interpretability of convolutional architectures. Its ability to model long-range dependencies is crucial for understanding ICH spatial distribution patterns.
The cascade architecture enables progressive feature refinement through a two-stage process:
📐 Cascade Architecture Diagram
Visualization showing EfficientNet-V2 → Feature Fusion → ConvNeXt → Classification
Combines local and global features for comprehensive hemorrhage detection across all sizes and locations
Two-stage refinement reduces false positives while maintaining high sensitivity for small hemorrhages
Cascade connections enable efficient feature sharing and gradient flow, improving training stability
Architecture naturally handles multiple concurrent hemorrhage types, reflecting clinical reality
The system employs three independently trained models to provide robust predictions and reduce uncertainty.
Trained with standard augmentation and balanced sampling
Trained with different initialization and augmentation strategies
Trained with emphasis on challenging cases and edge scenarios
256 × 256 × 3 (RGB)
6 (Multi-label Binary)
TensorFlow / Keras
Binary Cross-Entropy
Adam with Learning Rate Scheduling
32 (Training)