3D Knee MRI Cartilage Segmentation
Description
The Osteoarthritis Dataset consists of 3D knee MRI volumes acquired using the DESS (Dual Echo Steady State) sequence from the Osteoarthritis Initiative (OAI) study. Each volume has dimensions of 384×384×160 voxels with single-channel grayscale intensity values. The dataset is designed to support automated segmentation of knee joint structures, with particular focus on cartilage tissue identification and defect detection. Segmentation Classes The dataset includes ground truth annotations for five distinct classes: Background (Class 0): Non-cartilage tissue, synovial fluid, and surrounding anatomical structures Femoral Cartilage (Class 1): Cartilage tissue covering the femoral condyles Tibial Cartilage (Class 2): Cartilage tissue covering the tibial plateau Bone (Class 3): Femoral and tibial bone structures Defect (Class 4): Cartilage defects, lesions, and areas of degeneration Dataset Structure The dataset is organized into three main splits: Training Set: Approximately 70% of total scans, used for model training Validation Set: Approximately 15% of total scans, used for hyperparameter tuning and model selection Test Set: Approximately 15% of total scans, reserved for final model evaluation Each split contains image slices organized by class labels (0-4), facilitating class-balanced data loading and analysis. Image Characteristics Voxel Spacing: Approximately 0.3-0.5 mm isotropic resolution Field of View: Covers entire knee joint including femoral and tibial cartilage regions Image Quality: High signal-to-noise ratio with excellent cartilage-to-fluid contrast Intensity Range: Normalized grayscale values suitable for deep learning processing Format: NIfTI (.nii, .nii.gz) or PNG image slices
Files
Steps to reproduce
Data Preprocessing The dataset has undergone standardized preprocessing: Intensity normalization using percentile clipping (1st-99th percentile) Spatial normalization to consistent dimensions (384×384×160 voxels) Quality control validation of ground truth masks Format standardization for compatibility with deep learning frameworks Use Cases This dataset is designed for: Automated knee cartilage segmentation Cartilage defect detection and characterization Osteoarthritis progression monitoring Volume quantification for clinical assessment ICRS grading support systems Development of computer-aided diagnosis tools