Differential and relaxed image foresting transform for graph-cut segmentation of multiple 3D objects.

Area of Science:

• Medical Imaging
• Computer Vision
• Image Segmentation

Background:

• Interactive binary segmentation is time-consuming, especially for multiple objects.
• Existing graph-cut algorithms require significant user input for accurate delineation.

Purpose of the Study:

• To present DRIFT, an algorithm for efficient 3D multiple object segmentation.
• To enable interactive refinement of segmentation results using seed voxels.

Main Methods:

• DRIFT utilizes seed voxels and Differential Image Foresting Transforms (DIFTs) with relaxation.
• The algorithm incorporates diffusion filtering for boundary smoothing and seed-based connectivity correction.
• Iterative execution allows users to refine segmentation by adding or removing seed voxels.

Main Results:

• DRIFT achieves efficient segmentation, with initial runs in linear time and subsequent corrections in sublinear time.
• The algorithm was evaluated on 3D CT images of the thorax.
• Successful segmentation of multiple thoracic structures including arterial and venous systems, esophagus, pleural cavities, and trachea/bronchi was demonstrated.

Conclusions:

• DRIFT offers an efficient and interactive approach for 3D multiple object segmentation.
• The algorithm's iterative nature and seed-based refinement improve segmentation accuracy and user efficiency.
• DRIFT shows promise for segmenting complex anatomical structures in medical imaging.