Lung Lesion Extraction Using a Toboggan Based Growing Automatic Segmentation Approach.

Area of Science:

• Medical Imaging
• Radiology
• Computational Pathology

Background:

• Accurate lung lesion segmentation in computed tomography (CT) scans is critical for lung cancer research, clinical diagnosis, and treatment planning.
• The heterogeneity of lung lesions presents a significant challenge for fully automatic detection and segmentation with high accuracy.

Purpose of the Study:

• To introduce a novel, automated segmentation approach for lung lesions in CT images.
• To evaluate the performance of the proposed method in terms of detection sensitivity, segmentation accuracy, and efficiency.

Main Methods:

• A three-step framework termed toboggan based growing automatic segmentation (TBGA) was developed.
• The framework includes automatic initial seed point selection, multi-constraints 3D lesion extraction, and final lesion refinement.
• The method requires no human interaction or training datasets for lesion detection.

Main Results:

• The TBGA approach achieved a high lesion detection sensitivity of 96.35%.
• Segmentation accuracy was comparable to manual segmentation (P > 0.05) on both LIDC-IDRI and in-house datasets.
• TBGA demonstrated significant improvements in segmentation accuracy compared to level set and skeleton graph cut methods, with an average segmentation time under 8 seconds per lesion.

Conclusions:

• The novel toboggan based growing automatic segmentation (TBGA) method enables robust, efficient, and accurate automatic lung lesion segmentation in CT images.
• This automated approach has the potential to significantly aid lung cancer research and clinical practice.
• The TBGA method overcomes limitations of existing techniques by eliminating the need for manual intervention or training data.