TL;DR
This paper introduces a novel near-infrared spectroscopy (NIRS) catheter system for high-density mapping of radiofrequency ablation (RFA) lesions in the left atrium. The system demonstrates robust performance in blood and is insensitive to catheter angle, offering a promising tool for improving the safety and efficacy of atrial fibrillation treatments.
What Problem Does This Paper Address?
Atrial fibrillation (AF) is a common and costly cardiac arrhythmia. Catheter-based radiofrequency ablation (RFA) is a primary treatment, but its success hinges on creating complete and durable lesions. Current methods for assessing lesion quality, like electroanatomic mapping (EAM), have limitations in accurately characterizing lesion extent and depth, leading to recurrence rates that necessitate repeat procedures. While other imaging modalities exist, they often face challenges with cost, integration, or limited penetration. This paper addresses the need for a real-time, high-resolution, and robust method to map RFA lesions in situ, even in the presence of blood, to guide ablation procedures more effectively and reduce recurrence.
Methods
The researchers modified a commercial irrigated RFA catheter by integrating two multimode optical fibers for illumination and detection. This custom NIRS catheter was coupled with an electromagnetic tracking system to precisely record the catheter's position and orientation. The system acquires near-infrared reflectance spectra from the tissue surface.
Key methodological aspects include:
- Custom NIRS Catheter: Integration of optical fibers into an existing RFA catheter without compromising its ablation capabilities.
- Electromagnetic Tracking: Enables real-time, image-registered visualization of the catheter tip location and allows for dense tissue mapping with thousands of data points.
- Spectral Processing: Raw spectra are calibrated, converted to relative reflectance, and normalized. Specific optical indices (e.g., LOI, SOI) are calculated from these spectra, which are known to be sensitive to tissue changes caused by ablation.
- Machine Learning Classification: A random forest (RF) model was trained and tested to classify spectra as either lesion or non-lesion tissue. This model was chosen for its high classification accuracy.
- Ex Vivo Porcine Heart Model: Experiments were conducted on healthy pig hearts to simulate the ex vivo environment. Ablations were performed, and then mapping was done in both phosphate-buffered saline (PBS) and porcine blood to assess robustness.
Datasets
The study utilized data collected from 22 healthy pig hearts. For classification, the dataset was divided into a training set (32,392 spectra from 18 hearts) and a testing set (7124 spectra from 4 hearts). The training set comprised 87.9% non-lesion and 12.1% lesion spectra, while the testing set had 85.1% non-lesion and 14.9% lesion spectra.
Key Insights and Ideas
- Blood is Not a Barrier: A significant finding is that NIRS measurements and subsequent lesion classification remain highly accurate even when the tissue is submerged in blood. This is crucial for in vivo applications where blood is always present.
- Angle Insensitivity: The optical indices used for lesion detection showed remarkable insensitivity to catheter incidence angles up to approximately 60 degrees. This is a major advantage for mapping the complex geometry of the left atrium, where achieving perfect perpendicular contact is often difficult.
- High-Density Mapping: The combination of the NIRS catheter and electromagnetic tracking allows for the acquisition of thousands of data points, creating high-density lesion maps that can provide more detailed information than traditional EAM.
- Random Forest Dominance: Among the tested machine learning models, the random forest classifier demonstrated the highest accuracy in distinguishing between lesion and non-lesion tissue, highlighting its effectiveness for this application.
Key Results
- Classification Accuracy: The trained random forest model achieved a sensitivity of 81.7%, specificity of 86.1%, and an ROC AUC of 0.921 on the test set.
- Blood vs. PBS Equivalence: Quantitative comparison metrics (SSIM, inner product, RMS difference, and Dice coefficient) showed high similarity between lesion maps generated in PBS and blood, indicating that NIRS mapping is equivalent in both media. Mean SSIM was 0.749, and mean inner product was 0.867.
- Angle Tolerance: Optical index values remained distinct for lesion and non-lesion tissue even at contact angles up to 60 degrees, demonstrating robustness to oblique contact.
- Lesion Map Similarity: Predicted lesion maps achieved a mean structural similarity index of 0.749 and a mean normalized inner product of 0.867 when comparing maps obtained in PBS and blood.
Limitations and Future Work
- Pulmonary Vein Tissue Exclusion: Pulmonary venous tissue was excluded from analysis due to its thinness, which made it difficult to acquire spectra unaffected by the underlying corkboard. Future work could involve modifying dissection and mounting strategies.
- Room Temperature Mapping: While ablation was performed at 37°C, mapping was done at room temperature. Although bath temperature is not expected to affect optical measurements, this should be verified.
- Healthy vs. Diseased Hearts: The study used healthy pig atria, which do not fully replicate the pathological changes seen in AF patients. Future research should involve collecting NIRS data from diseased human hearts.
- In Vivo Validation: The study was conducted ex vivo. The authors emphasize the need for validation with in vivo large animal experiments to fully replicate a genuine RFA procedure.
- Optical Index Optimization: While several optical indices were effective, the authors suggest that dimensionality reduction techniques could be explored to identify optimal wavelengths for future studies.
Summary updated