Multi Texture Analysis of Colorectal Cancer Continuum Using Multispectral Imagery

Area of Science:

• Digital pathology
• Medical imaging analysis
• Cancer research

Background:

• Colorectal cancer (CRC) characterization is crucial for diagnosis and treatment.
• Multispectral optical microscopy offers detailed tissue visualization.
• Texture analysis can reveal subtle pathological differences.

Purpose of the Study:

• To characterize the colorectal cancer (CRC) continuum using texture features.
• To differentiate between benign hyperplasia, intraepithelial neoplasia, and carcinoma.
• To evaluate the efficacy of texture features in multispectral image analysis for CRC.

Main Methods:

• Active contour segmentation to extract regions of interest from multispectral images.
• Texture feature extraction using Laplacian-of-Gaussian (LoG) filters, discrete wavelets (DW), and Gray Level Co-occurrence Matrices (GLCM).
• Statistical analysis (Kruskal-Wallis test) and classifier models for evaluating feature significance and predictive ability.

Main Results:

• Significant texture differences (p < 0.01) were observed between pathological tissue types.
• Gray Level Co-occurrence Matrices (GLCM) features showed superior individual performance in predicting tissue types.
• Combining all texture features yielded a high classification accuracy of 98.92%, with 98.12% sensitivity and 99.67% specificity.

Conclusions:

• The combined approach using multiple texture features is effective for characterizing the colorectal cancer (CRC) continuum.
• This method demonstrates high efficiency in discriminating between various pathological tissues in multispectral images.
• Texture analysis of multispectral images provides a robust tool for pathological tissue classification in CRC.