Texture analysis for colorectal tumour biopsies using multispectral imagery.

Area of Science:

• Digital Pathology
• Medical Imaging Analysis
• Oncology

Background:

• Colorectal cancer diagnosis relies on histological analysis of biopsy samples.
• Multispectral imaging offers potential for capturing richer features from cancer tissues compared to traditional methods.
• High dimensionality and feature variability in multispectral data pose analytical challenges.

Purpose of the Study:

• To investigate and compare the performance of texture feature extraction techniques on multispectral colorectal biopsy images.
• To evaluate the efficacy of different classifiers for analyzing these complex datasets.
• To determine if multispectral imaging offers advantages over panchromatic approaches in colorectal cancer diagnosis.

Main Methods:

• Texture feature extraction using Local Binary Patterns (LBP), Haralick features, and Local Intensity Order Patterns (LIOP).
• Analysis of multispectral and panchromatic imagery from colorectal biopsy samples.
• Classification using Support Vector Machine (SVM) and Random Forest algorithms.

Main Results:

• Multispectral imaging demonstrated superior performance compared to the classical panchromatic approach.
• The combination of Local Binary Patterns (LBP) and Support Vector Machine (SVM) achieved the highest accuracy of 91.3%.
• Specific texture features extracted from multispectral data are highly discriminative for colorectal cancer.

Conclusions:

• Multispectral imaging is a valuable tool for enhancing colorectal cancer histological analysis.
• LBP texture features, when analyzed with SVM, provide a robust and accurate method for classifying colorectal cancer.
• This approach holds promise for improving the accuracy and efficiency of colorectal cancer diagnosis.