Breast Cancer Detection with Low-dimension Ordered Orthogonal Projection in Terahertz Imaging

Area of Science:

• Biomedical Imaging
• Data Science
• Cancer Research

Background:

• Terahertz (THz) imaging generates high-dimensional spectral data for biological tissues.
• Accurate cancer detection relies on effective analysis of complex THz spectral information.
• Existing dimension reduction methods may not fully capture unique features in THz breast cancer images.

Purpose of the Study:

• To introduce a novel dimension reduction algorithm, low-dimension ordered orthogonal projection (LOOP).
• To apply LOOP for improved feature extraction from THz images of human breast cancer tissues.
• To enhance the accuracy of cancer detection and classification using THz imaging data.

Main Methods:

• Development of the low-dimension ordered orthogonal projection (LOOP) algorithm for dimension reduction.
• Projection of high-dimensional THz spectral vectors into a low-dimension subspace.
• Utilizing a multivariate Gaussian mixture model with unsupervised learning (MCMC, EM) for statistical analysis.
• Classification of tumor regions based on extracted low-dimension features.

Main Results:

• The LOOP algorithm effectively reduces dimensionality while preserving unique spectral features.
• A significant performance improvement was observed in human breast cancer tissue analysis compared to existing methods.
• Classification results demonstrated a strong correlation with histopathology findings.

Conclusions:

• The proposed LOOP algorithm is a promising technique for breast cancer detection using THz imaging.
• Dimension reduction is crucial for extracting meaningful information from complex THz spectral data.
• This approach offers potential for more accurate and reliable cancer diagnosis.