Spectral Detector CT-Derived Pulmonary Perfusion Maps and Pulmonary Parenchyma Characteristics for the Semiautomated Classification of Pulmonary Hypertension

Objectives To evaluate the usefulness of spectral detector CT (SDCT)-derived pulmonary perfusion maps and pulmonary parenchyma characteristics for the semiautomated classification of pulmonary hypertension (PH). Methods A total of 162 consecutive patients with right heart catheter (RHC)-proven PH of different aetiologies as defined by the current ESC/ERS guidelines who underwent CT pulmonary angiography (CTPA) on SDCT and 20 patients with an invasive rule-out of PH were included in this retrospective study. Semiautomatic lung segmentation into normal and malperfused areas based on iodine density (ID) as well as automatic, virtual non-contrast-based emphysema quantification were performed. Corresponding volumes, histogram features and the ID SkewnessPerfDef-Emphysema-Index (δ-index) accounting for the ratio of ID distribution in malperfused lung areas and the proportion of emphysematous lung parenchyma were computed and compared between groups. Results Patients with PH showed a significantly greater extent of malperfused lung areas as well as stronger and more homogenous perfusion defects. In group 3 and 4 patients, ID skewness revealed a significantly more homogenous ID distribution in perfusion defects than in all other subgroups. The δ-index allowed for further subclassification of subgroups 3 and 4 (p < 0.001), identifying patients with chronic thromboembolic PH (CTEPH, subgroup 4) with high accuracy (AUC: 0.92, 95%-CI, 0.85–0.99). Conclusion Abnormal pulmonary perfusion in PH can be detected and quantified by semiautomated SDCT-based pulmonary perfusion maps. ID skewness in malperfused lung areas, and the δ-index allow for a classification of PH subgroups, identifying groups 3 and 4 patients with high accuracy, independent of reader expertise.


INTRODUCTION
Pulmonary hypertension (PH) describes a rare group of diseases that are defined by an increase in mean pulmonary artery pressure (mPAP) at rest as assessed by right heart catheterization (RHC). While the current ESC/ERS guidelines define PH as an increase in the mPAP ≥ 25 mmHg (1) the 6th World Symposium on Pulmonary Hypertension Task Force more recently suggested a new pressure level of > 20 mmHg to define an abnormal elevation in the mPAP (2). According to the ESC/ERS guidelines PH is clinically classified based on hemodynamic characteristics and the review of a comprehensive set of clinical investigations. The clinical classification defines five subgroups: pulmonary arterial hypertension (PAH, group 1), PH due to left heart disease (group 2), PH due to lung disease and/or hypoxia (group 3), chronic thromboembolic PH (CTEPH) and other pulmonary artery obstructions (group 4) and PH with unclear/multifactorial mechanisms (group 5). While all groups share an increase in the mPAP, the hemodynamic classification differentiates between precapillary PH, defined by a pulmonary arterial wedge pressure (PAWP) ≤ 15 mmHg and a concomitant elevation in pulmonary vascular resistance (PVR) ≥ 3 Wood Units (WU) (group 1, 3, 4, 5) and post-capillary PH (PAWP > 15 mmHg, PVR < 3 WU, group 2 and 5) (1, 2). Based on the diastolic pressure difference (DPG) group 2 is further subclassified into isolated precapillary PH (Ipc-PH, DPG < 7 mmHg, PVR < 3 WU) and combined pre-and postcapillary PH (Cpc-PH, DPG ≥ 7 mmHg and PVR >3 WU) (1). The underlying pathophysiological mechanisms that lead to loss and obstructive remodeling of the pulmonary vascular bed are diverse and remain incompletely understood. These range from sustained vasoconstriction, pulmonary vascular remodeling and in situ thrombosis, over heart failure, hypoxic vasoconstriction and Abbreviations: AUC, Area under the receiver operating characteristic curve; CI, Confidence interval; CO, Cardiac output; COPD, Chronic obstructive pulmonary disease; Cpc-PH, Combined pre-and postcapillary pulmonary hypertension; CTEPH, Chronic thromboembolic pulmonary hypertension; CTPA, CT pulmonary angiography; DECT, Dual energy CT; DPG, Diastolic pressure difference; ID, Iodine density; IDI, Iodine density image; Ipc-PH, Isolated postcapillary pulmonary hypertension; LA, Left atrium; MPA, Main pulmonary artery; mPAP, Mean pulmonary artery pressure; PAH, Pulmonary arterial hypertension; PCWP, Pulmonary capillary wedge pressure; PerfDef, Perfusiondeficit/malperfused lung areas; PerfNorm, Normalperfused lung areas; PH, Pulmonary hypertension; PVR, Pulmonary vascular resistance; RA, Right atrial; RHC, Right heart catheterization; RVEDD, Right ventricular end-diastolic diameter; SDCT, Spectral detector CT; sPAP, Systolic pulmonary artery pressure; SvO 2 , Pulmonary arterial oxygen saturation; TR velocity, Tricuspid regurgitation velocity; TTE, Transthoracic echocardiography; VNC, Virtual non-contrast; δindex, ID Skewness PerfDef -Emphysema-Index. an obliteration of the pulmonary vasculature to a prolonged occlusion of the pulmonary vascular bed (3).
As a result of the broad variety of underlying pathomechanisms and the overlapping presentation in RHC (elevated precapillary pressure in subgroups 1, 3, 4, and 5), a repertory of diagnostic tests (e.g., echocardiography, RHC, pulmonary function testing and blood gases, V/Q-scintigraphy) is recommended to achieve the final diagnosis in patients with suspected PH (1). Timely diagnosis is of crucial importance, as it not only defines therapy but also PH, regardless of the cause, is associated with poor prognosis (4)(5)(6)(7).
With the introduction of dual energy CT (DECT), mapping of pulmonary perfusion based on the different absorption characteristics of iodine and lung parenchyma has become available (8). The generated iodine density images (IDIs) are considered a sufficient surrogate parameter to estimate organ perfusion (9,10) and have proven to provide information on pulmonary perfusion in the setting of acute pulmonary embolism as well as in CTEPH with comparable accuracy to V/Qscintigraphy (8,(11)(12)(13)(14)(15)(16)(17)(18). In comparison to V/Q-scintigraphy, which remains the cornerstone to screen for CTEPH, DECT offers the advantage of allowing a comprehensive analysis of the lung parenchyma, pulmonary perfusion and vessel anatomy in a single examination (13,16,17,19). With regard to the complete spectrum of PH, current data suggest that V/Qscintigraphy (20) and DECT-based pulmonary perfusion maps might aid in the differentiation of PH subgroups, as different types of perfusion abnormalities correlate with PH etiology (21)(22)(23). In addition to the evaluation of pulmonary perfusion and vasculature in a "one-stop-examination, " DECT offers the unique possibility of quantifying parenchymal lung disease, a potential cause of PH, based on virtual non-contrast (VNC) images without the necessity of extra radiation exposure. Various studies indicate a comparable accuracy of VNC-based emphysema quantification compared to real non-contrast images (24,25), which have also been shown to discriminate PH due to lung disease and/or hypoxia from PAH (26). As opposed to the traditional imaging approach in PH with V/Q-scintigraphy as the central modality to differentiate between CTEPH and non-CTEPH and subsequent contrast or non-contrast CT imaging for further patient characterization the integrated DECT approach yields potential to serve patient needs as well as financial considerations (27).
Notwithstanding, according to the current guidelines, DECT only plays a supportive role in the diagnostic work-up of PH (1). This appears reasonable, as its diagnostic accuracy excessively depends on reader expertise, hampering its implementation in clinical routine especially in centers with limited expertise in FIGURE 1 | Study-flow-chart. PH, pulmonary hypertension; RHC, right heart catherization; CTPA, CT pulmonary angiography; Ipc-PH, precapillary pulmonary hypertension; Cpc-PH, combined pre-and postcapillary pulmonary hypertension.
PH imaging (23). In addition, current data are either limited due to the small sample size and/or rely on time-consuming manual image interpretation, inheriting the limitation of intraand interreader variability Further, hitherto most DECT studies have either assessed its ability to differentiate between health and disease or two distinct subgroups, but not the whole spectrum of PH (16,17,22,23).
A recently developed software application for volumetric iodine quantification enables iodine quantification per voxel for a 3D dataset acquired on a dual-layer CT platform (spectral detector CT, SDCT) (28). This allows for a semiautomatic, threshold-based segmentation of the lung into normal and malperfused areas based on iodine concentration.
Given the potential diagnostic merit of pulmonary perfusion in PH, the aim of our study was to evaluate whether semiautomatically derived volumetric parameters of SDCTbased pulmonary perfusion maps can aid the diagnosis and classification of PH.

Study Population
The study cohort consisted of 201 consecutive patients who had been referred to the Department of Cardiology of the University Hospital of Cologne for PH screening between May 2016 and May 2019 and underwent CT pulmonary angiography (CTPA) on SDCT. In accordance with the 2015 ESC/ERS guidelines, patients with suspected PH underwent RHC, ventilation/perfusion scintigraphy (V/Q scan) to exclude distal CTEPH and further testing (1). CTPA and RHC were routinely performed during the same hospitalization. Patients with suspected CTEPH had to be anticoagulated for three months in advance at the time of referral. Patients with detected hypoxia had to be on stable treatment for at least four weeks before RHC and CTPA examination. The final diagnosis was reached by expert consensus based on all available diagnostic tests, including CTPA. Given the retrospective design of this study final diagnosis was based on the 2015 ESC/ERS guidelines, defining PH as an increase in the mPAP ≥ 25 mmHg (1). Twenty of these 201 patients with an invasive rule-out of PH via RHC (mPAP < 25 mmHg) served as a control cohort (Figure 1). Among controls, besides a variety of cardiopulmonary diseases, the most frequent diagnoses were hypertensive cardiomyopathy (35.0%), chronic thromboembolic disease with no PH (15.0%) and systemic sclerosis (10.0%).
This retrospective study was approved by the local institutional review board (Ethics committee of the Faculty of Medicine from the University of Cologne, Cologne, Germany). Due to the retrospective design of the study the local institutional review board waived necessity for informed consent. All clinical investigations were conducted in accordance with the Declaration of Helsinki.

Image Acquisition and Reconstruction
CT data were acquired on a clinically available spectral detector CT (IQon, Philips Healthcare). All patients received an intravenous 50 ml bolus of contrast media (300 mg iodine/ml, Accupaque, GE Healthcare) followed by a 40 ml NaCl flush injected with a flow rate of 4 ml/s. Scanning was initiated with a delay of 4.9 s after an attenuation of 150 HU was reached in the main pulmonary artery (MPA). The acquisition parameters were as follows: slice collimation 64 × 0.625 mm; rotation time 0.33 s; tube potential 120 kV, automatic tube current modulation was used, reference mAs 75. For all reconstructions, a soft tissue kernel (B, Philips Healthcare) and a dedicated spectral reconstruction algorithm were used (Spectral, Philips Healthcare). Images were reconstructed in axial orientation with a slice thickness of 1 mm and a slice overlap of 0.5 mm. Matrix was set to 512 × 512. In addition to conventional images, which are identical to images reconstructed with the vendors hybriditerative reconstruction algorithm (29) (iDose4, Philips Healthcare), iodine maps and virtual non-contrast images were reconstructed.

SDCT-Based Pulmonary Perfusion
Automatic segmentation of the lung was performed using the commercially available software solution (Intellispace Portal, COPD tool, Version 10, Philips Healthcare). All segmentations were manually verified by a radiologist and in case of insufficient segmentation, lung volumes were manually edited. Subsequently, lung volumes were semiautomatically segmented into normal and malperfused areas via threshold segmentation based on the IDIs using dedicated software for volumetric iodine quantification (ISD, ThresholdSegmentation (1.1) Philips Intellispace Release 11). Three lung areas were defined as previously described (30), corresponding thresholds given brackets: malperfused (PerfDef): voxels with an iodine density (ID) of ≤5 % of the MPA (PerfDef = 0.0-≤5 % ID of the MPA), normal perfused (PerfNorm): voxels with an ID of more than 5% of the MPA and ≤50 % of the left atrium (LA) (PerfNorm ≥5 % MPA-≤50 % LA), vessel compartment (Ves): voxels with an ID of more than 50% of the LA (Ves = > 50 % MPA−100). Mean IDs in the MPA and LA were measured by manually drawing a ROI that accounted for at least 50% of the vascular/atrial area at the largest diameter. The following parameters were computed from histogram analysis for normal and malperfused lung areas: mean ID, ID kurtosis and ID skewness. Values for mean ID are given as standardized values calculated by division by the mean ID of the feeding vessel (MPA), as described previously (31). Figure 2 visualizes the segmentation workflow. Figure 3 illustrates the effect of ID distribution on ID skewness.

Lung Emphysema Quantification
Lung emphysema was quantified based on VNC images using a commercially available software solution (Intellispace Portal, COPD tool, Version 10, Philips Healthcare). The tool makes use of a thresholdbased emphysema calculation approach applying a threshold of −950 Hounsfield units (HU). Insufficiently segmented lung volumes were manually reedited (Supplementary Figure 1).

ID Skewness PerfDef -Emphysema-Index
The ID skewness in malperfused lung areas (ID Skewness PerfDef )-Emphysema-Index (δ-index) reflecting the ratio of ID distribution in malperfused lung areas and the proportion of emphysematous lung parenchyma was calculated forming the quotient of these two parameters. Discrete translation was performed to the parameter ID Skewness PerfDef by more than  the smallest negative measured value (−1.35) to ensure that the data distribution was transformed in a strictly positive range. To obtain a calculable devisor, the proportion of emphysematous   SDCT, spectral detector CT; AUC, area under the receiver operating characteristic curve; CI, confidence interval; ID, iodine density. * at a set sensitivity equal to normal perfused lung volume.
lung parenchyma was also transformed.
= ID Skewness PerfDef + 1.36 Proportion of emphysematous lung parenchyma + 1.0 The δ-index aims at distinguishing between homogenous pulmonary perfusion defects caused by parenchymal lung alterations and perfusion defects caused by true vascular malperfusion, e.g., due to vascular narrowing and/or obstruction as seen in CTEPH. While in both cases the homogenous ID distribution within the perfusion defect leads to an ID skewness close to zero (Figure 3), the index will either in-or decrease, depending on the amount of emphysematous lung alterations. Table 1 gives an overview over all parameters as assessed from RHC and SDCT.

Statistical Analysis
Normal distribution was tested using the Shapiro-Wilk test. Differences in continuous, parametric data were compared using the t-test. Continuous, independent, non-parametric data were compared using the Mann-Whitney U test. Differences in categorical data were identified using Pearson's chi-squared test. Variances among and between the subgroups concerning continuous data were compared using ANOVA for parametric data. After assessing the equality of variances using Leven's test, post hoc testing was performed by Bonferroni adjustment for multiple comparisons. On the other hand, differences between the subgroups for categorial, non-parametric variables were assessed using the Kruskal-Wallis test and Dunn-Bonferroni corrected post hoc analysis. The area under the receiver operating characteristic curve (AUC) was calculated for subclassification. Based on AUC analysis parameters, sensitivity and specificity for subclassification were calculated using Youden's index. A p-value of < 0.05 was considered statistically significant. Statistical analysis was performed using SPSS software (IBM SPSS Statistics for macOS, Version 27.0, Armonk, NY, USA).

Study Population
All patient demographics are given in Table 2 and  Supplementary Table 1. 201 patients were screened for PH. Of these, three patients were not eligible for RHC, and 14 had an inconclusive RHC or too low pulmonary arterial contrast for image analyses, defined by an ID in the MPA < 5 mg/mL. Subgroup 5 was excluded because of its small sample size (n = 2). In 20 patients, RHC ruled out PH (mPAP < 25 mmHg), and these patients served as controls (Figure 1). Table 3 displays all SDCT-based lung perfusion parameters in normal and malperfused lung areas for patients with PH and controls. Based on the ID in the MPA and the LA, right-toleft-heart contrast transit via the pulmonary vasculature was similar between groups. Patients suffering from PH showed higher proportions of malperfused lung areas (p = 0.02) as well as stronger (mean ID, p = 0.03) and more homogenous perfusion deficits (ID skewness, p = 0.03). Neither ID skewness in normal perfused lung areas nor ID kurtosis in any lung compartment differed significantly between groups. Based on AUC analysis the proportion of normal perfused lung areas qualified as the best SDCT-parameter to identify patients with PH (AUC: 0.67, 95%-CI, 0.56-0.78, sensitivity, 74%, specificity 55%, applying a cut-off of 71,6% as determined by Youden's index), yielding comparable diagnostic accuracy to the NT-pro BNP (AUC: 0.69, 95%-CI, 0.57-0.80, sensitivity, 59%, specificity 80%). At balanced sensitivities for both parameters the proportion of normal perfused lung areas was more specific than the NT-pro BNP (55 vs. 50%) ( Table 4).

SDCT-Based Differentiation of PH Subgroups
The Kruskal-Wallis test revealed significant differences between PH subgroups for the percentage of normal perfused lung volume (V PerfNorm ) and malperfused lung volume (V PerfDef ), mean ID, ID kurtosis as well as ID skewness in malperfused lung areas (p for all <0.001) ( Table 5). The comparison for all baseline characteristics between PH subgroups is given in Supplementary Table 2.
Patients suffering from PH due to chronic lung disease and/or hypoxia had the largest extent of V PerfDef among all subgroups, with a significantly higher proportion of malperfused lung areas compared to patients with Ipc-PH (Table 5). Moreover, malperfused lung areas showed a significantly stronger (mean ID) and more homogenous perfusion deficit (ID skewness) compared to all other subgroups (p for all < 0.001, Figure 4), except for subgroup 4 (p = 1.00, respectively).
Corresponding results were demonstrated for patients with CTEPH showing a significantly stronger perfusion deficit as well as more homogenous ID distribution in malperfused lung areas compared to all other subgroups (p for all < 0.001, only for mean ID 4-6 p = 0.01, Figure 4), except subgroup 3 (p = 1.00). ID skewness in malperfused lung areas allowed for an accurate identification of subgroups 3 and 4 based on AUC analysis (AUC: 0.79, 95%-CI, 0.72-0.86). Applying a cut-off of −0.335 as determined by Youden's index, a sensitivity of 84% and a specificity of 70% could be achieved for subclassification. Figure 5 illustrates the difference in pattern and ID distribution of malperfused lung areas in patients with PAH and patients suffering from CTEPH.
Neither normal nor malperfused lung areas revealed any difference between group 1 and patients suffering from Ipc-PH or Cpc-PH.
Excluding group 3 the prevalence of co-existing lung disease varied between 25.0% and 40.0% among subgroups ( Table 5). 8 patients (33.3%) from subgroup 3 were suffering from parenchymal lung disease other than chronic obstructive pulmonary disease (COPD). Patients from subgroup 3 with COPD did not show a higher amount of emphysematous lung alterations as assessed based on VNC reconstructions than those with lung fibrosis (p = 0.28). The δ-index differed significantly between subgroups 3 and 4 (Table 5; Figure 6). Based on AUC analysis, the index allowed for an accurate identification of patients with CTEPH (AUC: 0.92, 95%-CI, 0.85-0.99). A cutoff of 0.502 as determined by Youden's index was best for subclassification (sensitivity 86%, specificity 85%). The diagnostic accuracy of the δ-index was not affected by the presence of lung fibrosis (AUC: 0.92, 95%-CI, 0.84-1.00). The stepwise differentiation of both subgroups based on IDIs and VNC-based lung emphysema quantification is exemplified in Figure 7.

DISCUSSION
With the to our knowledge, the first study evaluating the potential of SDCT-derived pulmonary perfusion maps and pulmonary parenchyma characteristics for the semiautomated diagnosis and classification of PH with differing etiologies we can present several notable findings. (1) Semiautomatic lung segmentation into normal and malperfused lung areas based on ID values is feasible and applicable in the setting of PH. (2) SDCT-based FIGURE 5 | IDIs with automatically derived malperfused lung areas (red) and corresponding ID histograms of the perfusion deficit volume illustrating the difference of pattern and ID distribution of malperfused lung areas in two example cases of PAH (A,B) and CTEPH (C,D), respectively. Corresponding histograms indicate an inhomogeneous ID distribution in PAH (ID skewness < 0) and a more homogenous ID distribution in CTEPH (ID skewness close to 0). Thous., thousand; ID, iodine density. pulmonary parameters can not only detect but also quantify pulmonary perfusion abnormalities in PH holding comparable accuracy for PH diagnosis as NT-pro BNP. (3) ID skewness in malperfused lung areas and the introduced δ-index could potentially yield merit for a semiautomatic identification of PH subgroups, especially for the subclassification of groups 3 and 4.

Differentiation of PH Patients and Controls
The accurate diagnosis of PH remains challenging (33) and no meaningful decrease in the time from symptom onset to diagnosis could be achieved throughout the last decades (26). This also accounts for the high morbidity and mortality of PH, regardless of the underlying etiology (4)(5)(6)(7). Although RHC is accompanied by peri-and postinterventional complications and does not allow for morphologic information (34), it remains the diagnostic gold standard (1), as non-invasive imaging parameters either lack diagnostic accuracy (33,35,36) or have not yet been validated in large study cohorts, albeit promising initial results (37,38). This is in line with our findings. Although our data demonstrate that SDCT can not only assess but also quantify pulmonary malperfusion in PH, the diagnostic implications of these findings seem to be limited: while the volume of normal perfused lung areas yielded comparable diagnostic accuracy to the NT-pro BNP and thus might aid the detection of disease, the overall diagnostic accuracy of the SDCT-based pulmonary perfusion parameters was only moderate. Moreover, subgroup analysis revealed no significant difference in any volumetric pulmonary perfusion parameter between group 1 and Ipc-PH and controls (p for all > 0.05, data not shown). Concurrently, Kim et al. found only 60.0%, and Giordano et al. even found only 52.6% of PAH patients showing abnormal pulmonary perfusion (21,22). Although it remains unclear to what extent our results might be affected by the composition of the control group, which consisted of patients with the clinical suspicion of PH, a variety of cardio-pulmonary diseases and a borderline mPAP of 21.0 mmHg, our data thus underscore the challenges of a non-invasive PH diagnosis.

Classification of PH Subgroups
Our results demonstrate the merit of SDCT-derived pulmonary perfusion maps and pulmonary parenchyma characteristics over and above RHC for the classification of PH subgroups, particularly the spectrum of precapillary PH, which remains the main task from the radiologist's view (22).
There is evidence from V/Q-scintigraphic and DECT-based studies that pulmonary perfusion maps might enable the differentiation of PH subgroups, as different types of perfusion abnormalities correlate with PH etiology (20-23). Regardless of the modality, these studies are limited due to small sample sizes and/or time-consuming quantitative or semiquantitative image interpretation with considerable intra-and interreader variability. We demonstrated that semiautomatically generated volumetric parameters of SDCT-based pulmonary perfusion maps can aid the classification of PH. On the basis of ID skewness in malperfused lung areas, patients suffering from PH due to chronic lung disease and/or hypoxia and patients with CTEPH could confidently be differentiated from all other subgroups. Corresponding to the characteristic multisegmental, sharply defined, wedge-shaped and hypoattenuated appearance (17,39), which is also described for PE (12) and is pathophysiologically caused by either acute (PE) or prolonged (CTEPH) occlusion of the pulmonary arteries (3,40), malperfused lung areas in patients with CTEPH stood out due to more pronounced and more homogenous perfusion defects. The same was true for patients FIGURE 7 | Stepwise approach for the identification and differentiation of groups 3 (A1,A2) and 4 (B1,B2) based on ID skewness in malperfused lung areas, VNC-based quantification of emphysematous lung parenchyma and the δ-index. IDI, iodine-density image; ID, iodine density; V PerfDef , malperfused lung volume; thous., thousand; (ID Skewsness PerfDef , ID skewness in malperfused lung areas; VNC, virtual non-contrast, δ-index, ID Skewness PerfDef -Emphysema-Index. with PH due to chronic lung disease and/or hypoxia, which can be explained by the emphysematous lung changes prevalent in this patient group and the consecutive hypoxic vasoconstriction and obliteration of the pulmonary vascular bed (3) leading to large areas with no iodine attenuation at all. On the contrary pulmonary perfusion defects in groups 1 and 2 revealed a more heterogenous ID distribution as compared to groups 3 and 4, reflecting the more diffuse and less regional nature of pulmonary vascular remodeling in these groups.
Together with subgroups 1 and 5, subgroups 3 and 4 form the spectrum of precapillary PH. The subclassification of precapillary PH marks a key point in the diagnostic workup of PH, as the distinct forms cannot be differentiated via RHC (1). In this context, the identification of patients with CTEPH plays a pivotal role since these patients face a particularly poor prognosis (41) and can potentially be treated by surgical thrombendarteriectomy or balloon angioplasty. The evaluation of DECT-derived pulmonary perfusion maps for the diagnosis of CTEPH has been validated in numerous studies against V/Qscintigraphy (16,17), which is considered the gold standard to screen for the disease (1). With the introduction of the SDCT-based δ-index, we were able to identify a powerful parameter that allows for the semiautomated identification of patients with CTEPH. Contrary to previous SDCT studies differentiating CTEPH from other forms of PH (16,17,23), this approach works independently of reader expertise. This might be helpful to facilitate confident diagnosis of CTEPH for radiologists with limited expertise in PH imaging. In comparison to V/Q-scintigraphy, it overcomes the modality's inherent main limitation of V/Q-scintigraphy, which is unable to obtain morphologic information such as parenchymal changes or vessel anatomy (13,16,17,19). In combination with the interpretation of the pulmonary vasculature, SDCT-based pulmonary perfusion maps might thus aid the identification of patients with CTEPH and a high probability of profiting from treatment (22). Unlike a sequential approach of V/Q-scintigraphy and conventional CTA, the integrated SDCT approach offers the unique possibility of quantifying parenchymal lung disease based on VNC images without the necessity of extra radiation exposure (24,42). The δ-index thus automatically accounts for the radiologist's task to interpret iodine maps in correlation with pulmonary parenchyma alterations to differentiate between true perfusion defects and perfusion defects due to pulmonary pathologies such as emphysema (41).

Limitations
Apart from the retrospective study design and the missing validation of our findings, several limitations need to be addressed. First, the strict indication for an RHC intervention led to a considerable selection bias of the control group, which did not consist of truly healthy individuals but patients with a clinical suspicion of PH, a variety of cardio-pulmonary diseases and a borderline mPAP. Noteworthy, only 8 of 20 controls had a mPAP ≤ 20 mmHg. On the other hand, the invasive characterization of the controls is a strength of this study. Notwithstanding, a validation of our findings in controls with an mPAP < 20 mmHg is highly desirable, especially in the light of the upcoming change in the hemodynamic definition of PH (2). Second, the inability of the δ-index to differentiate between true perfusion defects and pseudodefects, e.g., due to beam hardening or motion artifacts, is a considerable limitation (43). Third, the δ-index does not allow for a direct quantification of other parenchymal lung disease than lung emphysema, e.g., lung fibrosis. Its diagnostic accuracy thus indirectly depends on the co-existence of emphysematous and non-emphysematous parenchyma alterations. Worth mentioning, in group 3 there was no significant difference between patients with and patients without lung fibrosis regarding the amount of emphysematous lung alterations, which is in line with the increasingly recognized coexistence of both diseases (44,45). Further, the presence of lung fibrosis did not have a negative influence on the diagnostic accuracy of the δ-index for the differentiation between groups 3 and 4. Fourth, due to scan timing, SDCT-derived perfusion maps only provide very limited information on lung compartments with maintained blood supply beyond the occluded pulmonary arteries through systemic collateral blood flow, e.g., via bronchial arteries, best described for CTEPH (41), and thus might overestimate the degree of perfusion deficit. This probably represents the methods biggest limitation. Fifth, we did not differentiate between major-vessel and minor-vessel CTEPH, which is known to partly mimic typical PAH perfusion deficit patterns (22). Validation of our findings in a PAH and subclassified CTEPH population thus seems highly desirable. Last, our data did not allow for an independent validation of the VNC-based emphysema quantification, as the CTPA protocol did not include true non-contrast images. Further studies addressing this considerable limitation are highly warranted.

Conclusion
SDCT-derived pulmonary perfusion and pulmonary parenchyma characteristics can detect and quantify pulmonary perfusion abnormalities in PH and allow for a semiautomated diagnosis of groups 3 and 4, independent of reader expertise.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by Ethics Committee of the Faculty of Medicine from the University of Cologne, Cologne, Germany. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.