Diagnostic value of transmural perfusion ratio derived from dynamic CT-based myocardial perfusion imaging for the detection of haemodynamically relevant coronary artery stenosis

The local institutional review board approved this prospective study. Written informed consent was obtained from all patients. This study included cases from a previous study investigating the diagnostic performance of CT-MPI [7]. Patients with suspected or known CAD referred for invasive angiography were prospectively recruited. Included patients underwent a dynamic CT-MPI examination 1-14 days before invasive angiography.

This study was designed to investigate the ability of CT-MPI to detect ischaemia; therefore only territories with an FFR measurement in the associated coronary artery were included. Territories associated with a (sub)total occluded coronary artery where no FFR measurement could be performed were not included in the analysis.

Patients with suspected or known coronary artery disease referred for invasive angiography were recruited in the time period December 2010 until December 2014. Exclusion criteria were younger than 40 years old, impaired renal function (serum creatinine >120 μmol/l), possible pregnancy or breast feeding, body weight over 120 kg, use of clopidogrel, contra-indications for iodine contrast medium, or contra-indications for adenosine.

All patients were requested to refrain from caffeine intake 24 h prior to the examination. In both arms 18-gauge cannulas were inserted in the antecubital veins. Blood pressure and ECG were monitored during the examination. Forty patients were scanned with a second-generation dual-source CT scanner and three patients with a third-generation dual-source CT scanner (SOMATOM Definition Flash and SOMATOM Force, Siemens Medical Solutions, Forchheim, Germany). Adenosine was infused at a rate of 140 μg/kg/min. CT-MPI acquisition was started 3 min after start of adenosine infusion.

The acquisition protocol consisted of coronary CT angiography, a non-contrast scan and the dynamic CT-MPI scan. The non-contrast scan was acquired during end systole and served for planning of the CT-MPI. Before the CT-MPI acquisition all patients received sublingual nitroglycerine. Intravenous beta blockers were used in patients with high heart rates prior to the coronary CT angiography, but very infrequently (N = 3) as these potentially affect the CT-MPI performance. After 3 min of adenosine infusion, 50 ml of contrast medium (Ultravist, 370 mgI/ml; Bayer, Berlin, Germany) was injected at 6 ml/s, followed by a saline bolus of 40 ml. All CT-MPI studies were made with an axial scan mode at 250 ms after the R wave (end systolic). To sufficiently cover the left ventricle the myocardial acquisition was performed in alternating cranial and caudal table positions (shuttle mode), acquiring two slightly overlapping data sets [3]. CT-MPI acquisition was started 5 s after the start of the contrast medium injection. Patients were asked to hold their breath during the entire dynamic CT-MPI acquisition (30-35 s). The number of time points acquired varied per patient depending on the heart rate: 1 patient had 9 cranial and caudal time points, 11 patients had 10, 6 patients had 11, 12 patients had 12, 9 patients had 13, and 4 patients had 14 time points.

The second-generation dual-source CT scanner used the following scan parameters: collimation 2 × 64 × 0.6-mm detector collimation with flying z-spot technique [11], gantry rotation time 280 ms, temporal resolution 75 ms, tube voltage/current 100 kVp/300 mAs and shuttle-mode coverage 73 mm.

The third-generation dual-source CT scanner used the following scan parameters: collimation 2 × 96 × 0.6-mm detector collimation with flying z-spot technique, gantry rotation time 250 ms and temporal resolution 66 ms; Care-Kv [12] was used with reference settings for tube voltage/current: 80 kVp/300 mAs and shuttle-mode coverage 102 mm.

The CT-MPI images were reconstructed using a dedicated kernel for reduction of iodine beam-hardening artefacts (b23f, Qr36) and transferred to a CT-MPI analysis software package (Volume Perfusion CT body, Syngo Somaris/7; Siemens, Germany). Motion correction was applied if necessary to correct for breathing displacement. The motion correction algorithm uses a time point selected by the user (with contrast in the left and right ventricle and smooth connection between the cranial and caudal section) and then registers the other time points to the selected time point using non-rigid registration. The left ventricle is segmented by combining thresholding and peak enhancement [13]. The change of attenuation in the myocardium over time was computed by creating time-attenuation curves (TACs). For quantification of the MBF the influx of contrast bolus was measured with an arterial input function (AIF). The AIF was measured by placement of an ROI in the descending aorta in the CT-MPI images. Precision of the AIF was increased by including both the cranial and caudal sections (double sampling). For quantification of the MBF the myocardial TACs were coupled with the AIF using a hybrid deconvolution model. The model generates a perfusion model curve based on the change in attenuation using a simplified impulse residue function for modelling the interaction between the intra- and extracellular compartments. The MBF was computed on a per voxel basis by dividing the maximal slope of the model curve for the myocardial tissue by the maximum AIF [3, 4, 14]. MBF data sets were reconstructed with a 512 × 512 matrix resulting in a pixel size of 0.35 × 0.35 mm and were reconstructed as a stack of colour-coded maps with a slice thickness of 3 mm and an increment of 1.5 mm.

MBF and TPR were individually evaluated by readers with previous experience in dynamic CT-MPI examinations. Both readers were provided with the colour-coded CT-MPI data sets. For each patient a list of vessels investigated by FFR was provided. To ensure correct territory-vessel correspondence left or right coronary dominance was provided for each patient. All readers were asked to measure MBF or TPR value corresponding to the vessel where the FFR measurements were performed. Each independent reader was blinded to all other medical information.

Within the MBF short axis slice interpreted as representing the myocardium dependent on the vessel in which the FFR was made, a freehand ROI was placed surrounding the suspected perfusion defect (Syngo Via 2.0, Siemens AG, Germany). The freehand ROI had a minimal area of 50 mm². Careful considerations were made to prevent inclusion of artefacts in the ROI.

For TPR the CT-MPI colour-coded maps were visually assessed to identify the slice most representative for a subendocardial/subepicardial ratio. The section of interest was loaded onto a dedicated image analysis application (ImageJ 1.48, National Institutes of Health, USA) [15]. To measure the transmural differences in the MBF a series of linear samples perpendicular to the myocardial surface was taken at 0.4-mm equal intervals (Fig. 1). The mean MBF values from the pixels under the line are projected in the transmural MBF profile curve. Care was taken not to sample too close to the LV lumen and epicardial border as the MBF absolute values are unreliable because of displacement artefacts. From the short-axis MBF image and the MBF profile curve the user selected the endocardial and epicardial positions. The TPR was calculated by dividing the subendocardial by the subepicardial MBF.

Invasive coronary angiography was performed according to local clinical standards. Prior to the invasive angiography intracoronary nitroglycerine was given, as is the standard in our centre. Invasive FFR was performed in all vessels with a visual stenosis grade between 30-90 % by invasive angiography. By protocol, an FFR pressure wire (PressureWire Aeris/Certus, St. Jude Medical, St. Paul, USA, or Prime/Combo Wire, Volcano, San Diego, CA, USA) was placed distal to the stenosis of interest, after which hyperaemia was induced by intravenous infusion of adenosine at 140 μg/kg/min. An invasive FFR ≤0.80 was considered haemodynamically significant.

Absolute variables are represented as total and percentage, continuous variables as mean and standard deviation (±). The mean values for MBF and TPR for normal and ischaemic territories were compared with an unpaired two-sided independent t-test. Pearson coefficient correlation was calculated for respectively MBF and TPR against invasive FFR. The receiver-operator characteristic (ROC) curves including the area under the curve (AUC) were presented for MBF and TPR. To investigate the combined diagnostic performance for MBF and TPR an ROC curve was also plotted for a new combined variable MBF multiplied by TPR (MBF × TPR). The optimal threshold for MBF and TPR diagnostic accuracy was calculated using the Youden index [16]. A sub-analysis was made for territories with an intermediate MBF between 50-100 ml/100 ml/min, as these represent territories with MBF values close to the diagnostic threshold [5, 7]. Diagnostic performance was evaluated as sensitivity, specificity, positive predictive value, negative predictive value and accuracy, with their corresponding 95 % confidence intervals (CI). The 95 % confidence intervals were corrected for within-subject clustering of data using variance adjustment [17]. MBF and TPR were displayed against each other with territories classified as normal or ischaemic. Inter-observer variability was determined for 72 (75 %) randomly selected territories by intra-class correlation coefficient for absolute MBF and TPR; diagnostic classification was compared using kappa statistics. Results were reported on a per-territory and per-patient basis and in accordance with the STARD initiative (Standard for Reporting Diagnostic accuracy) [18]. Most statistical analyses were made using SPSS (version 21, IBM Corp., Armonk NY, USA), while MedCalc (version 13.0; MedCalc Software, Ostend, Belgium) was used to compare the AUCs by using the method of DeLong et al. [19].