Interactive lung segmentation in abnormal human and animal chest CT scans

Area of Science:

• Medical image analysis
• Radiology
• Computational pathology

Background:

• Automatic segmentation of anatomical structures in medical imaging can fail with gross pathology.
• Accurate segmentation is crucial for many medical image analysis tasks, including thoracic CT scans.
• Existing methods may struggle with complex cases, necessitating improved approaches.

Purpose of the Study:

• To develop a versatile, fast, and reliable interactive system for segmenting anatomical structures.
• Specifically, to apply this system for segmenting lungs in challenging thoracic computed tomography (CT) scans.
• To evaluate the performance of both supervised interactive and purely interactive segmentation approaches.

Main Methods:

• Volumetric thoracic CT scans were segmented into 3D volumes of interest (VOIs) based on voxel density.
• VOIs were automatically labeled as lung or non-lung tissue, with options for interactive or supervised interactive correction.
• The supervised interactive system allowed user correction and continuous retraining, while the interactive system required manual correction of all mislabeled VOIs.
• Both methods were tested on 32 CT scans from pigs, mice, and humans with pulmonary abnormalities.

Main Results:

• Supervised interactive lung segmentation averaged under 9 minutes of user interaction and 2 minutes of computation time.
• The interactive method required an average of 13 minutes of user interaction.
• Both methods achieved high accuracy, with an average Dice similarity coefficient of 0.933 compared to manual delineations.
• On average, only 2.0% (supervised) and 3.0% (interactive) of VOIs required relabeling.

Conclusions:

• Two fast and reliable methods for interactive lung segmentation in challenging chest CT images were developed.
• These systems do not require prior knowledge of the scans and are effective across a variety of thoracic CT datasets.
• The developed interactive tools offer efficient and accurate solutions for lung segmentation in complex pathological cases.