Optical coherence tomography: A potential tool for unsupervised prediction of treatment response for Port-Wine Stains

Summary 

Background

Treatment of Port-Wine Stains (PWS) suffers from the absence of a reliable real-time tool for monitoring a clinical endpoint. Response to treatment varies substantially according to blood vessel geometry. Even though optical coherence tomography (OCT) has been identified as a modality with potential to suit this need, it has not been introduced as a standard clinical monitoring tool. One reason could be that – although OCT acquires data in real-time – gigabyte data transfer, processing and communication to a clinician may impede the implementation as a clinical tool.

Objectives

We investigate whether an automated algorithm can address this problem.

Methods

Based on our understanding of pulsed dye laser treatment, we present the implementation of an unsupervised, real-time classification algorithm which uses principal components data reduction and linear discriminant analysis. We evaluate the algorithm using 96 synthesized test images and 7 clinical images.

Results

The synthesized images are classified correctly in 99.8%. The clinical images are classified correctly in 71.4%.

Conclusions

Principal components-fed linear discriminant analysis (PC-fed LDA) may be a valuable method to classify clinical images. Larger sampling numbers are required for a better training model. These results justify undertaking a study involving more patients and show that disease can be described as a function of available treatment options.

Keywords: Discriminant analysis, Optical coherence tomography, Port-Wine Stain