fastrad
Documentation
Introduction
Why fastrad?
Use Cases
Installation
Prerequisites
Installing from Source
Hardware Acceleration (CUDA)
cuCIM for GLSZM Acceleration
Quickstart: Evaluating a Lung Nodule
Prerequisites
Step 1: Load the Clinical Volume
Step 2: Configure the Extractor Blueprint
Step 3: Execute the Extraction
Step 4: Consume the Feature Output
Next Steps
Advanced: Hardware-Accelerated Voxel-wise Feature Extraction
User Guide
Core Philosophy: PyTorch over Sequences
Object-Orientated Architecture
Extraction Workflow Execution
VRAM Protection and Fallbacks
Handling Anisotropy
Single Voxel Limitations
Learn (Examples)
01: Basic Feature Extraction
02: PyTorch CUDA GPU Acceleration Benchmark
03: Multi-Patient Clinical Batch Processing
04: Advanced Configurations & Fallback Logic
Features & Mathematical Formulations
First Order Statistics
Shape (2D and 3D)
Gray Level Co-occurrence Matrix (GLCM)
Gray Level Size Zone Matrix (GLSZM)
Gray Level Dependence Matrix (GLDM)
Neighbourhood Gray Tone Difference Matrix (NGTDM)
Benchmarks
IBSI Compliance and Numerical Parity
GPU Performance
CPU Performance
Multi-threading Fairness Benchmark
ROI Size Scaling Benchmark (GPU)
Optimizing GLSZM (cuCIM)
Stability Guarantee
ICC Analysis on Real RIDER Scan-Rescan Pairs
Numerical Robustness to Input Perturbation
Memory Footprint Optimization
GPU VRAM Profile (Full Pipeline)
Edge Case Handling
Dense Voxel-Wise Hardware Extraction Performance
API Reference
MedicalImage
Mask
FeatureSettings
FeatureExtractor
DenseFeatureExtractor
Frequently Asked Questions (FAQ)
Feature Extraction: Input, Customization, and Reproducibility
Hardware and Errors
Common Exceptions
Developers Guide
Architecture: PyTorch over SimpleITK
Adding the Baseline (Feature Classes)
Scientific Parity Testing
Out-Of-Memory (OOM) GPU Catchers
Contributing to fastrad
The PR Process and Gotchas
Submitting Bug Reports
fastrad
Index
Index