A novel pipeline for adrenal tumour segmentation

Area of Science:

• Medical Imaging
• Radiology
• Computational Pathology

Background:

• Adrenal tumours present segmentation challenges due to their location, varied shapes, sizes, and low contrast with surrounding tissues.
• Existing segmentation methods struggle with unilateral, bilateral, adherent, or non-cohesive adrenal tumours, lacking a comprehensive pipeline.

Purpose of the Study:

• To introduce an advanced adrenal tumour segmentation (ATUS) pipeline to address limitations in current adrenal tumour segmentation techniques.
• To develop a robust method for accurate segmentation of adrenal tumours, irrespective of their characteristics or relationship with adjacent organs.

Main Methods:

• The proposed ATUS pipeline integrates multiple image processing techniques: contrast limited adaptive histogram equalization, quadtree-based split and merge, mean shift segmentation, large grey level elimination, and region growing.
• This multi-method approach aims to enhance image contrast, refine segmentation boundaries, and consolidate regions for precise tumour delineation.

Main Results:

• The ATUS pipeline demonstrated significant performance across six metrics (Dice, Jaccard, sensitivity, specificity, accuracy, SSIM) on 32 CT images.
• Achieved high success rates (e.g., 83.06% Dice, 71.44% Jaccard, 86.44% sensitivity) even with tumours adhering to other organs.
• Performance was robust against challenges like size variation, diverse locations, and low contrast, indicating broad applicability.

Conclusions:

• The developed ATUS system provides detailed and accurate adrenal tumour segmentation.
• It effectively overcomes common obstacles in adrenal tumour segmentation, offering a reliable tool for clinical applications.