Lung segmentation from CT with severe pathologies using anatomical constraints

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Radiology

Background:

• Automatic lung segmentation in CT scans is difficult due to diverse appearances of diseased lung tissue.
• Severe pathologies present significant challenges for standard segmentation algorithms.
• Accurate lung segmentation is crucial for quantitative analysis and diagnosis.

Purpose of the Study:

• To develop and validate an algorithm for robust lung segmentation in CT images with severe pathologies.
• To improve the accuracy and reliability of automatic lung segmentation by incorporating contextual information.
• To reduce the rate of failed lung detections in challenging clinical cases.

Main Methods:

• A novel algorithm combining statistical learning with contextual constraints from neighboring anatomies (heart, liver, spleen, ribs).
• Utilizing anatomical relationships to guide and refine lung tissue segmentation.
• Testing the algorithm on unseen CT cases with a variety of severe pathologies.

Main Results:

• The proposed algorithm significantly reduces the number of failed lung detections.
• Improved accuracy in lung segmentation was observed on test cases with severe pathologies.
• The method demonstrated consistency with adjacent anatomical structures.

Conclusions:

• Incorporating contextual anatomical constraints enhances the robustness of lung segmentation algorithms.
• This approach offers a promising solution for segmenting lungs in complex CT imaging scenarios.
• The algorithm shows potential for improving diagnostic capabilities in thoracic imaging.