Application of unsupervised learning to hyperspectral imaging of cardiac ablation lesions

Abstract

Atrial fibrillation is the most common cardiac arrhythmia. It is being effectively treated using the radiofrequency ablation (RFA) procedure, which destroys culprit tissue and creates scars that prevent the spread of abnormal electrical activity. Long-term success of RFA could be improved further if ablation lesions can be directly visualized during the surgery. We have shown that autofluorescence-based hyperspectral imaging (aHSI) can help to identify lesions based on spectral unmixing. We show that use of k-means clustering, an unsupervised learning method, is capable of detecting RFA lesions without a priori knowledge of the lesions' spectral characteristics. We also show that the number of spectral bands required for successful lesion identification can be significantly reduced, enabling the use of increased spectral bandwidth. Together, these findings can help with clinical implementation of a percutaneous aHSI catheter, since by reducing the number of spectral bands one can reduce hypercube acquisition and processing times, and by increasing the spectral width of individual bands one can collect more photons. The latter is of critical importance in low-light applications such as intracardiac aHSI. The ultimate goal of our studies is to help improve clinical outcomes for atrial fibrillation patients.

Key factors behind autofluorescence changes caused by ablation of cardiac tissue

“…To date, we have confirmed that Auf-HSI can reveal ablation lesions made in the left atrium of large mammals, including pigs, sheep, and cows 9 , as well as in donated human heart tissue 10 . The dimensions and the shape of the lesions delineated by Auf-HSI were in perfect agreement with the conventional post-ablation staining methods such as TTC [10][11][12][13][14] . These promising bench findings have led to our ongoing efforts to incorporate Auf-HSI technology into a percutaneous imaging catheter 11,15,16 .…”

“…The important practical implication of our studies is the dramatically reduced amount of spectral information needed to reveal the ablated tissue. In the past, we have attempted to approach this issue mathematically 12 , 42 . In those studies, we used combinatorial tools to screen all possible wavelength combinations from the Auf-HSI datasets for a minimal number of wavelengths that can still provide good-quality outcomes.…”

A Percutaneous Catheter for In Vivo Hyperspectral Imaging of Cardiac Tissue: Challenges, Solutions and Future Directions

“…That is why we use term 'hyperspectral' when describing our findings. Our recent post-acquisition analysis of continuous spectra from these hyperspectral datasets revealed the most informative wavelengths, allowing us to consider use of fewer and wider bands [30,38]. In order not to confuse our readers by juggling between the two terms while referring to our previous publications and bench datasets, we decided to continue to refer to this technology as hyperspectral, although the final iteration of the clinical device is most likely to be classified as 'multispectral'.…”

Toward cardiac tissue characterization using machine learning and light-scattering spectroscopy

“…[16][17][18] Previously, our research group demonstrated the ability of a machine learning approach, i.e., a convolutional neural network (CNN), to detect fibrosis in FCM images taken in a beating heart in situ. 19,20 Other previous studies explored machine learning for cardiac ablation lesion quantification 21 and diagnosis of ischemic heart disease. 17,22 Here, we introduced a novel system for broad-spectrum LSS for applications in cardiology and cardiac surgery.…”