Evaluation of CTCA in Assessing Plaque Pathology and Physiology

STUDY DESIGN

1. Patent recruitment Seventy patients with typical angina symptoms who had elective coronary angiography showing at least one complex (i.e., bifurcation lesion, long lesion, calcified lesion) obstructive lesion (>70% diameter stenosis on coronary angiography, or a fractional flow reserve <0.80) that is considered suitable for percutaneous coronary intervention (PCI) under IVUS guidance will be included in the study. Exclusion criteria are: 1) age >75 years, 2) ACS within <3 months, 2) eGFR <60ml/min/1.73m², 3) previous coronary artery bypass surgery, 3) decompensated heart failure, or left ventricular ejection fraction ≤30%, 4) intravenous contrast allergy or inability to receive treatment with aspirin, heparin, or thienopyridines, 5) anticipated life expectancy <1 year, 6) history of heart transplantation, 7) patient that requires surgical revascularization, and 8) extensive coronary artery disease (i.e., multiple chronic total occlusions) or tortuous coronary anatomy that does not allow assessment of the coronary arteries with NIRS-IVUS imaging. The recruited patients will provide informed consent and undergo CTCA imaging using a dedicated 3rd generation CT dual-source scanner (Siemens Force). The first 4 months of the study effort will be made to optimise image acquisition protocols so as the obtained CTCA data to be suitable for automated segmentation. Efficient imaging matrixes will be used to improve in-plane spatial resolution and sharper reconstruction kernels and iterative reconstruction algorithms will be implemented to enhance image segmentation. Two weeks after CTCA imaging the patients will undergo planned PCI. During PCI effort will be made to study all the 3 epicardial coronary arteries - including the stenotic lesion - and some of their major side branches (i.e., large diagonals, obtuse marginals, the posterior descending artery or the left ventricular branch of the right coronary artery) with the combined NIRS-IVUS catheter. Following PCI the participants will be discharged on optimal medical treatment.
2. Segmentation of the CTCA imaging data and reconstruction of coronary artery anatomy Imaging data will be anonymised and analysed blinded to clinical details by an expert operator using dedicated workstation. Anatomical landmarks (i.e., side branches) will be identified in the CTCA and NIRS-IVUS imaging data and will be used to define segments of interest. Analysis of the CTCA data will be performed using dedicated software that enables automated extraction of the luminal centreline, semi-automated detection of the lumen and outer vessel wall borders, and quantification of the plaque burden and incorporates a plaque characterisation algorithm that allows automated characterisation of the composition of the plaque. The plaque characterisation algorithm takes into account predefined fixed intensity cut-off values of the Hounsfield units and an adaptive approach that allows modification of these cut-off values according to image attenuation. Currently the segmentation process takes on average 3h per patient. In this project the investigators aim to optimise CTCA image acquisition and segmentation algorithms so as this process to become automated and reduce the time for CTCA segmentation to <1 hour.
3. Segmentation of the NIRS-IVUS imaging data and reconstruction of coronary artery anatomy The NIRS-IVUS data portraying the segments of interest will be analysed by an expert operator, blinded to the clinical details and CTCA imaging data, with a 3-month interval from the analysis of the CTCA data using a software that enables detection of the lumen and outer vessel wall borders, quantification of plaque burden and annotation of the calcific tissue component in IVUS. The output of the analysis of the NIRS imaging data is the chemogram which is a colour coded map of the distribution of the lipid component along the vessel wall (yellow indicates increased probability and red low probability of lipid tissue). A metric of the lipid burden is the lipid core burden index (LCBI) which is computed as the fraction of the yellow pixels that correspond to lipid component divided by 1000. In addition, for each 2mm segments the block chemogram is generated that provides a summary of the chemogram for this segment and displays the probability of the presence of lipid tissue in a 2mm block of the coronary artery. The block chemogram has been validated against histology and it has been shown that it enables accurate detection of lipid-rich plaques. The segmented NIRS-IVUS data will be used to reconstruct the coronary anatomy using an established and well-validated methodology. Side branches with a diameter >1.5mm will be reconstructed from the angiographic data and fused with the main vessel geometry reconstructed from the NIRS-IVUS, since it has been shown that side branches affect ESS distribution.
4. Blood flow simulation Identical boundary conditions will be applied to both IVUS-based and CTCA-based models. Blood will be considered to be a laminar and incompressible Newtonian fluid with a dynamic viscosity of 0.0035 Pa•s and a density of 1,050 kg/m3. A steady flow profile will be imposed at the inflow of the lumen as this reduces computation time and there is evidence that there is no significant difference in the estimated ESS when a steady or a pulsatile flow profile is used. Murray's theory of constant ESS will be used to derive boundary conditions in the main and side branches. The arterial wall will be considered to be rigid and no-slip conditions will be applied at the luminal surface. Flow velocity will be estimated from the angiographic data by measuring the number of frames required for the contrast agent to pass from the inlet to the outlet of the reconstructed segment, the volume of the segment at baseline, and the cine frame rate.
5. Analysis of the NIRS-IVUS and CTCA imaging data It is anticipated that NIRS-IVUS imaging will be performed on average in 2.5 vessels per patient; from these 40 randomly selected vessels will be used to train the algorithms for CTCA segmentation and plaque characterisation (training dataset) and the remaining for validation purposes (validation dataset).

In the training set, the segments of interest reconstructed from the CTCA and NIRS-IVUS data will be divided in 2mm segments and corresponding 2mm segments will be identified in the CTCA and NIRS-IVUS models. For each 2mm segment the following metrics will be estimated in the NIRS-IVUS models: mean lumen area, mean outer vessel wall area, mean plaque area, mean plaque burden (defined as: 100 x plaque area/vessel area), mean calcific area, the LCBI and the predominant ESS. In addition each segment will be classified as lipid-rich or non-lipid rich according to the block chemogram.

Similarly, in the CTCA models the mean lumen area, outer vessel wall area, plaque area, plaque burden, calcific area and the mean predominant ESS will be estimated for every 2mm segment and compared with the estimations of NIRS-IVUS. Several approaches will be tested to optimise the segmentation of the vessel wall borders and the best will be adopted. Segments with increased calcific burden and blooming artifacts will be identified and in case of significant differences between CTCA and NIRS-IVUS annotations, machine learning techniques, that take advantage of the information provided by NIRS-IVUS, will be implemented to optimise CTCA segmentation. The adaptive Hounsfield unit cut-offs that best identify lipid and calcific tissue will be defined. Spread-out vessel plots portraying the distribution of the lipid tissue in the CTCA models will be created and in these the LCBICT will be estimated for each 2mm segment and compared with the output of NIRS. Area under the curve (AUC) analysis will be used to identify the best CT-derived plaque burden, LCBI and ESS cut-off values that correspond to the NIRS-IVUS cutoff values that indicate high-risk plaques (plaque burden: 67%, LCBI: 178 and ESS: 1Pa). The block chemogram in NIRS-IVUS will be used to identify the 2mm LCBICT cut-off that enables accurate classification of the 2mm segments in as lipid or non-lipid rich. The accuracy of these cut-offs will be tested in the validation dataset.

In addition, in the validation dataset the NIRS-IVUS data will be used to identify coronary lesions - defined as segments with a plaque burden >40% in 3 consecutive frames. For each lesion its remodelling index will be estimated and used to classify them as lesions with a positive or negative remodelling. The NIRS-IVUS data will be used to characterise their phenotype and classify them as: pathological intimal thickening/fibrotic plaques, fibro-calcific plaques, fibroatheromas (FA), and calcified fibroatheromas. The NIRS-IVUS lesion classification will be used as reference standard in order to assess the accuracy of CTCA in characterising lesion phenotype.

STATISTICAL ANALYSIS - POWER CALCULATION The primary endpoint of the study is the ability of CTCA in detecting FA. In a study of Garcia-Garcia that included 129 patients undergoing singe vessel IVUS imaging, 1.7 lesions were identified per patient. In the study of Puri et al., 45% of the lesions were FA on histology. In that study NIRS combined with IVUS enabled detection of FA with an excellent accuracy (c-index: 0.80). We anticipate that we will be able to perform NIRS-IVUS imaging in 2.5 coronary arteries per patient and that CTCA imaging quality will be optimal in 93% of the studied patients. If we recruit 70 patients we anticipate to successfully study with NIRS-IVUS and CTCA 162 vessels of which 120 (203 lesions - 92 FA) will be used as a validation dataset. This dataset is anticipated to give an 80% power to demonstrate using the 5% significance level, that the sensitivity of CTCA in identifying FA is not different from NIRS-IVUS (AUC of CTCA range: 0.89-0.71), assuming a true sensitivity of 0.80 for NIRS-IVUS.

Secondary endpoints of the study are the accuracy of CTCA to identify: a) lipid-rich segments (using the block chemogram of NIRS-IVUS as gold standard), and b) segments exposed to low ESS (<1Pa, using the ESS estimated in the NIRS-IVUS models as reference standard).