Using light-based catheters for radiofrequency ablation (RFA) therapies grants the ability to accurately derive tissue
properties such as lesion depth and overtreatment from spectroscopic information. However, this information is heavily
reliant on contact quality with the treatment area and the orientation of the catheter. Thus to improve assessments of
tissue properties, this work utilizes Bayesian modelling to classify whether the catheter is indeed in proper contact with
the tissue. Initially in-laboratory experiments were conducted with ten fresh swine hearts submerged in blood. A total of
1555 unique near infrared spectra were collected from a spectrometer using a light-based catheter and manually tagged
as “full perpendicular contact,” “angled contact,” and “no contact,” between the catheter and heart tissue. Three features
were prominent in all spectra for distinguishing purposes: area underneath the spectra, an intensity “valley” between 730
nm and 800 nm, along with the slope between 850 nm and 1150 nm. A classifier featuring bootstrapping, adaboost, and
k-means techniques was thus created and achieved a 96.05% accuracy in classifying full contact, 98.33% accuracy in
classifying angled contact, and 100% accuracy in classifying no contact.
|