Texture Feature Extraction and Analysis for Polyp Differentiation via Computed Tomography Colonography.

Area of Science:

• Medical Imaging
• Radiology
• Computational Pathology

Background:

• Image textures in computed tomography colonography (CTC) show promise for distinguishing between non-neoplastic and neoplastic polyps.
• Current CTC methods primarily focus on polyp detection, lacking diagnostic capabilities for polyp characterization.
• Texture analysis is often hindered by low-dose CT imaging and variations in polyp orientation, leading to inconsistent results.

Purpose of the Study:

• To develop an adaptive approach for robust texture feature extraction and analysis in CTC for improved polyp differentiation.
• To overcome limitations of texture analysis caused by low-dose imaging and polyp spatial orientation variations.
• To enhance the diagnostic capability of CTC beyond simple polyp detection.

Main Methods:

• Applied derivative operations (gradient, curvature) to CT intensity images to enhance textures while controlling noise.
• Utilized Haralick co-occurrence matrix (CM) to compute texture measures across 13 directions in intensity, gradient, and curvature images.
• Employed Karhunen-Loeve transform to create orientation-independent texture features from CM measures, reducing variability.

Main Results:

• The proposed adaptive texture analysis method demonstrated significant impact in polyp differentiation tasks.
• Experiments on 384 polyp datasets (52 non-neoplastic, 332 neoplastic) showed improved differentiation.
• Achieved an area under the curve (AUC) of 0.8016 in receiver operating characteristic analysis, indicating diagnostic feasibility.

Conclusions:

• The developed adaptive approach effectively enhances texture features and stabilizes their extraction against spatial variations in CTC.
• This method significantly improves the ability to differentiate between neoplastic and non-neoplastic polyps using CTC.
• The findings suggest that this technique holds diagnostic feasibility for advancing CTC applications.