Pulmonary nodules: improved detection with vascular segmentation and extraction with spiral CT. Work in progress

Area of Science:

• Medical imaging analysis
• Radiology
• Computer-aided diagnosis

Background:

• Pulmonary nodules are often difficult to detect on computed tomographic (CT) scans.
• Vascular structures can obscure or be mistaken for nodules, impacting diagnostic accuracy.

Purpose of the Study:

• To evaluate if automated pulmonary vessel extraction improves the detection rate of pulmonary nodules using CT imaging.
• To assess the impact of vessel segmentation on radiologist performance in identifying pulmonary nodules.

Main Methods:

• A three-dimensional seeded region-growing algorithm was used to extract pulmonary vessels from CT images.
• Simulated nodules were added to normal CT scans, and clinical data from eight patients were analyzed.
• Three radiologists assessed nodule detection on both original and vessel-extracted CT images, with performance measured by sensitivity and false-positive proportion.

Main Results:

• Pulmonary vessel extraction increased nodule detection sensitivity from 63% to 84% in simulations and 58% to 78% in clinical studies.
• The proportion of false-positive findings per case decreased substantially, from 52% to 24% (simulation) and 55% to 12% (clinical).
• Radiologists demonstrated significantly improved and consistent performance when evaluating CT images with segmented vessels compared to original images (P < .007).

Conclusions:

• Automated subtraction and extraction of pulmonary vessels from CT scans demonstrably enhance the accuracy of pulmonary nodule detection.
• This technique offers a valuable tool for improving diagnostic performance in radiology, particularly for identifying small or subtle pulmonary nodules.