Neural network and conventional classifiers for fluorescence-guided laser angioplasty

Area of Science:

• Biomedical Engineering
• Optical Diagnostics
• Cardiovascular Research

Background:

• Laser angioplasty aims to remove arterial plaque.
• Accurate tissue classification is crucial for effective and safe plaque ablation.
• Fluorescence spectroscopy offers a potential method for real-time tissue characterization.

Purpose of the Study:

• To investigate the efficacy of back-propagation and K-nearest neighbors algorithms in classifying arterial fluorescence spectra.
• To compare the performance of these machine learning techniques for differentiating atherosclerotic from normal arterial tissue.
• To infer characteristics of the classification problem based on algorithm performance variations.

Main Methods:

• Analysis of arterial fluorescence spectra using back-propagation neural networks.
• Application of K-nearest neighbors (KNN) algorithm for spectral classification.
• Comparative evaluation of classification performance between the two methods and other existing schemes.

Main Results:

• Both back-propagation and K-nearest neighbors techniques demonstrated competitive performance in classifying arterial fluorescence spectra.
• The study identified relative performance differences between variations of these algorithms.
• These performance variations provided insights into the geometric nature of the spectral classification task.

Conclusions:

• Back-propagation and K-nearest neighbors are viable methods for classifying arterial tissue using fluorescence spectra in laser angioplasty.
• These techniques can potentially guide selective ablation of atherosclerotic plaque.
• Further investigation into algorithm variations can refine optical diagnostic approaches for cardiovascular interventions.