Segmentation of extrapulmonary tuberculosis infection using modified automatic seeded region growing

Area of Science:

• Medical Imaging
• Image Segmentation
• Radiology

Background:

• Current image segmentation in PET/CT imaging primarily uses CT data, neglecting valuable PET information.
• Accurate image segmentation is crucial for developing automated diagnostic tools.

Purpose of the Study:

• To propose and evaluate a novel automatic image segmentation method for PET/CT scans.
• To leverage PET hot spot values to guide CT image segmentation using seeded region growing (SRG).

Main Methods:

• Implemented an automatic segmentation routine using seeded region growing (SRG).
• Introduced a new SRG growing criterion: sliding windows.
• Utilized PET hot spot values for initial seed selection and guiding CT segmentation.
• Evaluated performance on fourteen extrapulmonary tuberculosis patient images.

Main Results:

• SRG with local averaging and variance achieved the lowest over-segmentation percentage (2.67%).
• The modified SRG with local averaging and variance demonstrated the best performance in terms of time complexity (5.273s average execution time).
• The method showed good performance in reducing both over- and under-segmentation areas.

Conclusions:

• PET hot spot values effectively guide automatic CT image segmentation.
• The proposed modified SRG method enhances accuracy and efficiency in PET/CT image segmentation.
• This technique can be integrated into automated diagnostic systems for improved medical imaging analysis.