Texture analysis applied to second harmonic generation image data for ovarian cancer classification

Area of Science:

• Biomedical Engineering
• Computational Pathology
• Ovarian Cancer Research

Background:

• Extracellular matrix (ECM) remodeling is a key factor in ovarian cancer development.
• Quantifying ECM changes is crucial for accurate diagnosis and understanding disease progression.

Purpose of the Study:

• To develop and validate a texture analysis method for quantifying collagen fibrillar morphology in ovarian tissues.
• To differentiate between normal and high-grade malignant ovarian tissues using image analysis.

Main Methods:

• Utilized second harmonic generation (SHG) microscopy to image human ovarian tissues.
• Implemented a texture analysis approach creating a dictionary of "textons" (texture features).
• Developed a classification model based on texton distribution for normal and malignant tissues.

Main Results:

• Achieved up to 97% classification accuracy in distinguishing normal from high-grade malignant ovarian tissues.
• The local analysis algorithm effectively probes fibrillar morphologies.
• Optimized texton number and nearest neighbors for improved performance.

Conclusions:

• Texture analysis of SHG images provides a robust method for ovarian cancer detection.
• The developed algorithm offers a versatile and accurate approach to ECM remodeling analysis.
• This technique has potential for broader applications in disease state characterization.