Classification of terahertz-pulsed imaging data from excised breast tissue

Area of Science:

• Biomedical Engineering
• Medical Physics
• Computational Biology

Background:

• Terahertz (THz) pulse imaging offers potential for non-invasive tissue characterization.
• Accurate classification of breast tissue (tumor vs. normal) is crucial for diagnosis.
• Effective data analysis is needed to interpret complex THz pulse signals.

Purpose of the Study:

• To evaluate data reduction techniques for classifying terahertz pulse data from breast tissue.
• To assess the accuracy of support vector machine (SVM) classification with reduced data.
• To determine the optimal data reduction strategy for terahertz breast tissue analysis.

Main Methods:

• Studied 51 breast tissue samples (tumor and normal).
• Applied three data reduction methods: heuristic parameters, principal components of pulses, and principal components of parameter space.
• Utilized support vector machine (SVM) with a radial basis function for classification.

Main Results:

• Achieved a maximum classification accuracy of 92% using principal components of pulses and parameter space with ten components.
• Principal components of the parameter space showed superior performance when fewer than ten components were used.
• Demonstrated good classification on example images, acknowledging interpatient variability and edge effects.

Conclusions:

• Data reduction techniques combined with SVM classification are effective for accurate breast tissue classification using THz data.
• The chosen data reduction and classification methods show promise for clinical applications under controlled conditions.
• Further research may refine algorithms to address variability and improve robustness.