Department of Biomedical Engineering Posters and Presentations

Title

Toward Real-time Lesion Detection for Cardiac Ablation from Auto-fluorescence Hyperspectral Images

Document Type

Poster

Keywords

Atrial Fibrillation, Radiofrequency Ablation, Lesion Detection, Hyperspectral Imaging, Image Analysis

Publication Date

4-2017

Abstract

Atrial fibrillation (AF) is a common cardiac arrhythmia caused by abnormal electrical activity originating in ectopic foci in the left atrium. AF can be treated by creating scar tissue to eliminate that activity. Recently, radiofrequency ablation (RFA) has been used to treat the tissues, and electrical conduction testing is used to determine the ablated region. Electro-anatomical mapping can improve RFA therapy success rates. That method must visualize the ablated region directly during the procedure. MRI and CT had been proposed for visualization, but those methods are expensive, space-consuming, and do not provide real-time monitoring. This study seeks to make practicable another visualization approach called auto-fluorescence hyperspectral imaging (aHSI). In that approach, ultraviolet light illuminates the tissues in the left atrium and then tissues emit auto-fluorescence light. We wish to detect the loss of fluorescence of nicotinamide adenine dinucleotide (indicating loss of cells' viability, and thus successful ablation) even in the presence of the auto-fluorescence of atrial collagen. This requires detailed separation of the fluorescence spectra arising from successfully-ablated lesions and from unablated tissue. We use a camera that acquires aHSI data in bands (every 10nm from 420nm to 720nm) with a tunable filter. Light is delivered to and recorded from the cardiac tissue by fiber-optic light guides contained in a catheter. The goal of providing real-time identification of ablated tissue using low signal-to-noise-ratio (SNR) signals requires that the minimum number of bands of wavelengths be acquired and used in classification. A clustering method (k-means) was applied to detect the lesions in the aHSI images. Results show that the detected lesion areas are correct and clearly visible. Further study was conducted to decrease the number of bands from 31 to 4 (using bandpass filters) without reducing the accuracy of lesion detection. This assists in making the approach fast and robust to noise – an important consideration because the total light power that the catheter can deliver is limited. The power therefore should be allocated to the most useful wavelength bands. Through an optimization procedure, we divide the 31 original bands into four contiguous groups. We add the bands in each group to create new, increased-SNR images that are used to detect lesion areas by k-means. We find that four bands are sufficient for accurate lesion detection. These studies provide promising methods to realize real-time intraoperative monitoring and identification of ablated and unablated tissues during the procedure.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

Open Access

1

Comments

To be presented at GW Annual Research Days 2017.

This document is currently not available here.

Share

COinS
 

Toward Real-time Lesion Detection for Cardiac Ablation from Auto-fluorescence Hyperspectral Images

Atrial fibrillation (AF) is a common cardiac arrhythmia caused by abnormal electrical activity originating in ectopic foci in the left atrium. AF can be treated by creating scar tissue to eliminate that activity. Recently, radiofrequency ablation (RFA) has been used to treat the tissues, and electrical conduction testing is used to determine the ablated region. Electro-anatomical mapping can improve RFA therapy success rates. That method must visualize the ablated region directly during the procedure. MRI and CT had been proposed for visualization, but those methods are expensive, space-consuming, and do not provide real-time monitoring. This study seeks to make practicable another visualization approach called auto-fluorescence hyperspectral imaging (aHSI). In that approach, ultraviolet light illuminates the tissues in the left atrium and then tissues emit auto-fluorescence light. We wish to detect the loss of fluorescence of nicotinamide adenine dinucleotide (indicating loss of cells' viability, and thus successful ablation) even in the presence of the auto-fluorescence of atrial collagen. This requires detailed separation of the fluorescence spectra arising from successfully-ablated lesions and from unablated tissue. We use a camera that acquires aHSI data in bands (every 10nm from 420nm to 720nm) with a tunable filter. Light is delivered to and recorded from the cardiac tissue by fiber-optic light guides contained in a catheter. The goal of providing real-time identification of ablated tissue using low signal-to-noise-ratio (SNR) signals requires that the minimum number of bands of wavelengths be acquired and used in classification. A clustering method (k-means) was applied to detect the lesions in the aHSI images. Results show that the detected lesion areas are correct and clearly visible. Further study was conducted to decrease the number of bands from 31 to 4 (using bandpass filters) without reducing the accuracy of lesion detection. This assists in making the approach fast and robust to noise – an important consideration because the total light power that the catheter can deliver is limited. The power therefore should be allocated to the most useful wavelength bands. Through an optimization procedure, we divide the 31 original bands into four contiguous groups. We add the bands in each group to create new, increased-SNR images that are used to detect lesion areas by k-means. We find that four bands are sufficient for accurate lesion detection. These studies provide promising methods to realize real-time intraoperative monitoring and identification of ablated and unablated tissues during the procedure.