CT-ORG

CT-ORG provides a curated dataset and computational resources for multiple-organ segmentation in computed tomography (CT) to support training and evaluation of automated organ segmentation models.

Key Features:

• Dataset composition: 140 CT scans annotated with six organ classes: liver, lungs, bladder, kidney, bones, and brain.
• Diversity of cases: Collection includes a variety of cases intended to reflect a broad range of clinical imaging scenarios.
• Unsupervised morphological segmentation: Unsupervised morphological segmentation algorithms are used for annotating lungs and bones.
• 3D Fourier transform acceleration: The unsupervised morphological segmentation is accelerated using 3D Fourier transforms.
• Deep neural network segmentation: A deep neural network is trained to segment all six organs simultaneously, with reported runtime of 4.3 seconds per case.
• Data augmentation methodologies: Includes methodologies for efficient data augmentation to improve model generalization.
• GPU library: A GPU library is provided to facilitate the augmentation processes.

Scientific Applications:

• Model training: Training deep learning models for simultaneous segmentation of liver, lungs, bladder, kidney, bones, and brain in CT scans.
• Benchmarking and evaluation: Benchmarking and evaluating segmentation algorithms and architectures on a multi-organ CT dataset.
• Method development: Developing and validating unsupervised morphological segmentation approaches accelerated by 3D Fourier transforms.
• Augmentation research: Studying GPU-accelerated data augmentation strategies to enhance model robustness and generalization.

Methodology:

Unsupervised morphological segmentation algorithms accelerated by 3D Fourier transforms, deep neural network training for simultaneous six-organ segmentation, and GPU-enabled data augmentation.