Multifunctional intravascular optical coherence tomography for imaging inside arteries

Cardiovascular disease is the leading cause of death across the world.¹ In particular, plaque rupture (and subsequent complications in coronary arteries) is the main cause of myocardial infarction and heart attack. Furthermore, plaque rupture is strongly associated with the thickness of the fibrous tissue, the size of the lipid pool, inflammation, and the extent of macrophage infiltration.² As many of the underlying mechanisms of coronary artery disease occur at microscopic scales, high-resolution imaging techniques are urgently needed (i.e., to ‘look’ inside arteries) for research and diagnostic purposes.³ Intravascular optical coherence tomography (IVOCT) has unprecedented resolution (approximately 10μm), which is about 10 times higher than that of intravascular ultrasound,⁴ and has thus emerged as a useful clinical tool for visualizing microscopic details of arterial inner walls and of implanted stents in the coronary artery. Although IVOCT holds great promise for identifying and characterizing coronary plaques, i.e., by enabling the visualization of the microstructure of plaque (including the thin fibrous cap, lipid pool, and microphages),⁵ intrinsic artifacts of the technique (e.g., shadowing, tangential signal dropout, and lipid attenuation) mean that only limited-accuracy characterizations are possible.⁶ Moreover, manual visual assessment of grayscale IVOCT images can give rise to high inter- and intra-observer variability.

To date, many approaches have been developed to obtain accurate and robust quantitative analyses for the characterization of plaques. For example, the optical attenuation coefficient,⁷ backscattering coefficient,⁸ and textural features⁹ can be used. In addition, multimodal IVOCT systems—combined with near-IR fluorescence³ or near-IR spectroscopy¹⁰—have been introduced to provide additional information about the observed plaque. Another elegant approach is to obtain complementary information from the IVOCT signal by applying an additional spectroscopic analysis. Typically, IVOCT images are acquired by a Fourier transform of the interference signal between the reference and sample arm (using a broadband light source with a center wavelength of 1300nm and a bandwidth of about 100nm). This means that the IVOCT signal already contains spectroscopic information, which can be acquired from a short-time Fourier transform (see Figure 1).^6,11 The spectroscopic absorption property—extracted from the IVOCT signal—can thus be used to better characterize the plaque.¹¹

Figure 1. Illustration of the concept for conventional/general Fourier domain (FD) optical coherence tomography (OCT) and the proposed spectroscopic OCT methods. In the conventional OCT technique, a grayscale OCT intensity image is acquired via Fourier transform of the interferometric signal and is used to assess tissue morphology. However, if a short-time Fourier transform is applied to the interferometric signal instead of a standard Fourier transform (as in the spectroscopic OCT approach), additional spectroscopic information (i.e., depth-resolved spectra) can be obtained.⁶ λ: Wavelength. t: Time. k: Wavenumber.

Building on this previous work, we have developed a new method (based on spectroscopic analysis) for characterization and identification of lipid-rich plaques.⁶ For our algorithm, we use a specific metric—known as the Gaussian center of mass (GCOM)—which reflects the absorbance properties of lipids. We have also validated our approach using a lipid phantom and have thus shown that it can be used to overcome the normal limitations of grayscale IVOCT images.

Within the bandwidth of our IVOCT system, cholesterol-containing lipids have higher absorption properties at lower wavelengths, whereas other tissues exhibit rather flat absorbance through the bandwidth. In regions of lipid-rich plaque, the intensity at higher wavelengths is therefore much stronger than at lower wavelengths. This change in spectral shape can be expressed with a simple representative metric, such as the center of mass (COM). The COM, however, is likely to be distorted by speckle noise in an IVOCT signal. To accurately report the presence of lipid, we thus introduced our more robust GCOM metric. We calculate this term by applying Gaussian fitting to the IVOCT interference signal. We also calculate the derivative of the GCOM—the lipid distribution function (d_lipid)—to account for the cumulative effect of the tissue absorption in the depth direction.

To validate our methodology we used lipid phantoms (i.e., various concentrations of mayonnaise). Our results show that the mean d_lipid value is linearly correlated with the lipid concentration. We also successfully performed in vivo imaging of coronary-sized rabbit arteries. We show representative results from these experiments in Figure 2, in which the first, second, and third rows are results from lipid-rich plaque, fibrous plaque, and a normal artery, respectively. In addition, the first, second, and third columns show grayscale IVOCT images, d_lipid mapped onto IVOCT images, and the corresponding histological lipid (i.e., oil red O or ORO) staining, respectively. From these images we observed that in the lipid-rich plaque, the region with high d_lipid—marked by the yellow arrowheads in Figure 2(b)—corresponds well to the lipid in the ORO staining. In contrast, we observe low d_lipid values in the fibrous plaque and the normal artery: see Figure 2(e) and (h). We have also conducted a statistical analysis, the results of which support our in vivo findings and demonstrate that we can detect lipids with high accuracy (i.e., sensitivity of 94.3% and specificity of 76.7%).

Figure 2. Representative results from in vivo intravascular OCT imaging of coronary-sized rabbit arteries, and comparison with corresponding histology sections. Images show results for arteries with (a–c) lipid-rich plaque, (d–f) fibrous plaque, and (g–i) no lesions. Grayscale OCT data (a, d, g) only provides cross-sectional images, whereas the lipid distribution function (d_lipid)—derived from the spectral analysis—provides information about the presence of lipid. The d_lipid results (b, e, h) correspond well with the oil red O lipid staining (c, f, i). Yellow arrowheads mark regions with high d_lipid values. Scale bars indicate 500μm.⁶

In summary, we have presented a novel analytic method for the characterization of lipid-rich plaques from IVOCT images. We have also successfully demonstrated this technique—which is based on the spectroscopic absorption properties of lipids—with an atherosclerotic rabbit model. In particular, the attractiveness of our approach arises from our calculation of the lipid distribution function (derived from the conventional IVOCT signal, without the need for any additional devices). We are currently working on translating this technology into the clinic. In addition, we plan to further improve our algorithm so that other plaque components (e.g., microcalcifications and macrophages) can be detected. We expect that the complementary compositional information provided by our proposed method will be useful for overcoming the limitations of standard grayscale IVOCT images, and will thus provide a better assessment of high-risk plaque.