Pulmonary quantitative CT imaging in focal and diffuse disease: current research and clinical applications

Abstract

The frenetic development of imaging technology—both hardware and software—provides exceptional potential for investigation of the lung. In the last two decades, CT was exploited for detailed characterization of pulmonary structures and description of respiratory disease. The introduction of volumetric acquisition allowed increasingly sophisticated analysis of CT data by means of computerized algorithm, namely quantitative CT (QCT). Hundreds of thousands of CTs have been analysed for characterization of focal and diffuse disease of the lung. Several QCT metrics were developed and tested against clinical, functional and prognostic descriptors. Computer-aided detection of nodules, textural analysis of focal lesions, densitometric analysis and airway segmentation in obstructive pulmonary disease and textural analysis in interstitial lung disease are the major chapters of this discipline. The validation of QCT metrics for specific clinical and investigational needs prompted the translation of such metrics from research field to patient care. The present review summarizes the state of the art of QCT in both focal and diffuse lung disease, including a dedicated discussion about application of QCT metrics as parameters for clinical care and outcomes in clinical trials.

Introduction

The modern CT scanner has excellent spatial and temporal resolution for anatomical evaluation in vivo. The ability to derive quantitative CT (QCT) imaging provides a non-invasive mean for direct visualization, characterization and quantification of anatomic structures, as well as for speculation about pathophysiological processes of pulmonary diseases.¹

There is clearly interest in optimizing the role of QCT for application of its objective metrics, for instance in clinical trials and lung cancer screening.^2–4 Indeed, QCT techniques for characterization of interstitial lung disease (ILD), chronic obstructive pulmonary disease (COPD), as well as focal lesions (e.g. lung nodules), have been progressively developed, validated and refined over the past 20 years.

The purpose of the present review article is to summarize major QCT advances in both focal findings and diffuse lung disease, either in research or routine realms.

Detection and characterization of pulmonary nodule

State of the art

Computer-aided detection (CAD) of pulmonary nodule was progressively introduced, improved and validated for optimization of radiologist reading.⁵ In particular, CAD has been massively tested within lung cancer screening trials, where it could be optimized for both sensitivity and, notably, positive-predictive value (e.g. reduction of false-positive CAD findings is a major issue for optimal efficiency of the system).^6–8 CAD can act as first, second or concurrent-reader.⁹ The second reader approach shows the highest sensitivity at the cost of a greater reading time compared to the concurrent reader.^9–16

CAD has been evaluated for the detection of both solid and subsolid nodules.¹⁷ Sensitivity of CAD for solid nodules ranged from 38 to 100%.^18–22 Sensitivity for the detection of subsolid nodules of the first CAD software was poor²³ and currently shows a wide range between 54%²⁴ and above 70%.^{17, 25,26} This high heterogeneity derives from the technical differencesf the CAD systems commercialized by various manufacturers as well as from different studies’ methodologies.^{14, 27} Subsolid nodules are less frequent than the solid ones, nevertheless, they show higher incidence of lung neoplasms,²⁸ hence, it is mandatory that CAD be validated before its implementation in this subset of nodules.

CAD provides the radiologist with semi-automatic metrics driving the nodule management, such as volume and volume doubling time, which are more accurate than manual diameter.^{29, 30} Noteworthy, volumetric measurements and volume doubling time are particularly useful for the evaluation of size change during follow-up.³¹ Nevertheless, it cannot be overemphasized that variation occurs also for semi-automatic volumetry of nodules by CAD, notably as a function of size³² and in association with inspiratory effort.³³ It was proposed that 25% increase be an accurate threshold to define nodule growth in the follow-up of indeterminate solid nodules surrounded by aerated parenchyma (more limitations apply to nodules abutting vessels, bronchi and pleura).^{34, 35} A combination of volume and density is used to calculate the mass and the mass doubling time of subsolid nodules (Figure 1).³⁶ However, such metrics are not widespread,³⁷ partly because of the variability between different CAD software (even between different versions of the same software) that hampers longitudinal reproducibility (e.g. when measurements are performed in different radiology departments during follow-up).^{32, 38} The suggestion is to analyse the complete data set of CT time points with the same software at the time of the most recent CT scan, eventually with its most recent software version.

Figure 1. .

(a–i) CAD segmentation of subsolid nodule during a 4-year active surveillance. The semi-automatic segmentation of pulmonary nodule provides several metrics that can be used for standardized characterization and management. Noteworthy, the longitudinal assessment of subsolid nodule takes advantage of volumetric measurement of density, which was proposed for optimal stratification of nodule growth, also known as mass doubling time. From right to left, the same non-solid nodule segmented at baseline (a–c), after 2 years (middle column: d–f), and after 4 years (g–i) with progressive increase in growth rate according to MDT. For each time point the segmentation is rendered in axial (a, d and f), coronal (b, e and h) and sagittal plane (c, f and i). The MDT of this non-solid nodule rose from 1562 days at the 2-year LDCT to 350 days at the 4-year LDCT, reflecting a progressive increase in growth rate. CAD, computer-aided detection; LDCT, low-dose CT; MDT, mass doubling time.

Clinical application

CAD aims to reduce false-negative scans,^{13, 39,40} which are associated with the lone visual reading,⁴¹ thus improving sensitivity.^{14, 42} For instance, in a series of 400 low-dose CTs with 151 true-positive findings, 5 were missed by CAD while 33 by the two reading radiologists.²¹ However, in the current clinical daily practice, CAD are not commonly used because of the drawback of a high rate of false-positive findings requiring radiologists’ interaction^{5, 43,44} and the need for data transfer to a dedicated workstation. Noteworthy, lung cancer trials showed the usefulness of CAD for the semi-automated detection and measurement of nodules.^{6, 8,45} In keeping with the results of such large studies, it is believed that CAD will progressively be integrated also in clinical practice, increasing the sensitivity of radiologists for lung nodules.

Application in scientific investigation and clinical trials

The major issue of nodule management comes with the large amount of clinically silent lesions (e.g. benign, premalignant or malignant with indolent behaviour). Interobserver variability is the main limitation in objective assessment of nodule risk, yet, there are computerized alternatives for such task, also called radiomics. Radiomics is the high-throughput extraction of image features imperceptible for the human eyes,⁴⁶ such as pixel values, variation of those values within a region of interest (ROI) and edge strengths.⁴⁷

Currently, there are a number of software available for texture analysis (TA), each of them capable to extrapolate and evaluate different groups of radiomic features.⁴⁸ Radiomics works in multistep fashion composed of subsequent processes, namely: image acquisition, delineation of ROIs and extraction and analysis of features.⁴⁹ Noteworthy, features can be extracted with a semi-automated method reducing the number of manual inputs⁵⁰ and this approach might be beneficial if TA will be available in the clinical workflow.

Radiomic features extraction currently suffers from a significant inter-reader variability related to the selection of ROI.⁵¹ TA was shown to be capable of predicting patient outcome based on CT data sets acquired with different scanning parameters⁵²—the latter representing a frequent work setting— however, there is evidence about significant variability among CT data sets reconstructed with different algorithms.^{51, 53,54} Hence, future studies should investigate whether different scanning protocols can reliably be used for the TA-based stratification of patients. The consistency of radiomic features extracted from CT data sets obtained by different scanners will be particularly beneficial for clinical application as radiologists are frequently asked to compare CT studies acquired from different hospitals.

TA was evaluated to determine if radiomic features could differentiate between lung cancer and benign nodules,⁵⁵ as well as between transient and persistent part-solid nodules⁵⁶ or pre-invasive and invasive part-solid nodules.^{57, 58} In patients suffering from lung cancer, TA could non-invasively monitor changes in tumour heterogeneity in the early phase of therapy⁵⁹ to potentially provide objective biomarkers^{31, 60} for prediction of treatment response and survival.^{30,46,51,61–66} Furthermore, TA is increasingly investigated to stratify the risk of distant metastases from lung cancer⁵² as well as to provide in vivo non-invasive differentiation between histological types of lung cancer,⁶⁷ we hope future developments will supply robust metrics for this purpose.

Obstructive pulmonary disease

The heterogeneous framework of obstructive pulmonary disease can be biased on pulmonary function tests (PFTs). Conversely, quantitative imaging can detail abnormalities and differentiate between parenchymal and bronchial disease,^{68, 69} yet its validation for clinical application can be further demonstrated. CT is not recommended as part of the routine evaluation of obstructive pulmonary disease, nevertheless it can be employed in the phenotypization of chronic obstructive pulmonary disease (COPD),⁴ therapeutic planning for emphysema,⁷⁰ characterization of asthma,⁷¹ early detection of bronchiolitis obliterans syndrome after transplantation (e.g. bone marrow, lung)⁷² and even in paediatric obstructive diseases (e.g. cystic fibrosis, bronchopulmonary dysplasia).^{73, 74} The characterization of parenchyma and airway in obstructive pulmonary diseases has been anatomically and functionally covered by volumetric and multiphase CT acquisition.^{75, 76}

State of the art

Parenchymal quantification

Pulmonary texture in obstructive lung disease shows relatively low tightness compared to normal lung because of air abundance. Air abundance can be caused either by tissue loss (e.g. emphysema) or by functional limitation to air outflow (e.g. air trapping), oftentimes by a combination of them. The more the air the lower the pulmonary density, which can be quantified by volumetric segmentation of the lung and arithmetic analysis of density histogram (Figure 2). Densitometric quantification includes absolute thresholding and relative distribution of voxel density.

Figure 2. .

(a–d) Density histogram for computation of parenchymal metrics. Axial CT images of a patient with upper lobe predominant emphysema (a, b). The density histogram (c) summarizes the distribution of parenchymal density (white line: both lungs). Dedicated representation of individual lobar density histograms provides quantitative differences between lobes (red line: right upper lobe; green line: right lower lobe) for objective assessment of emphysema heterogeneity and selection of the most appropriate treatment (e.g. endobronchial valves vs endobronchial coils). Numeric output is automatically computed by the software (d) and displayed according to whole lung characteristics or according to the selected lobes (e.g. right and left lung or individual lobes). Quantitative metrics that can be extracted from the histogram include lung volume, mean lung density, weight, density percentiles (in the present table, the lowest 15th percentile is reported, P15), density mode (the most represented density value within the entire lung, Pk) and volume of lung parenchyma with density below a predefined threshold (in the present table, the relative volume of lung with density < −950 HU is reported as a measure of emphysema, LAV, low attenuation volume). HU, Hounsfield unit; MLD, mean lung density; Vol. volume; Wt., weight.

Pulmonary density is highly influenced by the respiratory phase;⁷⁷ this characteristic offers the opportunity for morphofunctional quantification of the lung.

On inspiratory scans, emphysema is conventionally attributed to areas of lung with density < −950 Hounsfield unit,⁷⁸ moreover, it is indirectly related to the lowest 15th percentile (P15)⁷⁹ (Figure 2). It cannot be overemphasized that lung density is also a function of tissue inflammation as it was shown in former smokers.⁸⁰ Therefore, lung density should always be interpreted within the appropriate pathophysiological context. It was proposed that emphysema quantification be integrated with lung mass for comprehensive depiction of a multiphase disease that might begin with active inflammation (higher density) and evolve towards irreversible tissue depletion (lower density). Washko reported that lung mass increasingly dropped from GOLD 1 to GOLD 4 (global initiative for chronic obstructive lung disease), and it predicted FEV1 (forced expiratory volume) decline in a large population of smokers (current or former).⁸¹

Volumetric expiratory scan for assessment of lung parenchyma is acquired at the end of expiration. Such acquisition was relatively challenging in the past, yet, it is currently made much easier by scanners with large arrays of detectors (e.g. 64 or more) that allow fast scan of the whole chest (scan time <5 s). Expiratory scan is useful to assess functional parenchymal change caused by obstructive small airway disease, namely air trapping. The densitometric parameters for quantification of air trapping include the relative area of lung < −856 Hounsfield unit on expiratory scan⁸² and the ratio between expiratory and inspiratory mean lung density.⁸³ However, these two metrics have potential bias due to the heterogeneity of parenchymal densitometry at different lung volumes, notably because air trapping and emphysema may be admixed.⁸⁴ Volumetric non-rigid registration of inspiratory and expiratory scan overcomes such limitation because it allows biphase characterization of each voxel, and is particularly interesting for quantification of air trapping. This approach was named parametric response mapping (PRM) when used to define density clusters for topographic categorization of parenchyma into normal lung, air trapping and emphysema⁸⁵ (Figure 3). A similar registration was used to simulate the local dynamic volume change. Bodduluri reported biomechanical CT descriptors of air trapping in association with patient outcome in a large COPD population.⁸⁶ Nonetheless, it should be realized that there is a normal range of air trapping even in young healthy subjects, which should be acknowledged for clinical application of QCT.^{87, 88}

Figure 3. (a-d) Parametric response mapping for topographic densitometric categorization of parenchyma into normal lung, air trapping and emphysema in COPD. Volumetric inspiratory (a) and expiratory (b) CTs are warped together for quantitative analysis of densitometric clusters by means of parametric response map (c): Insp > −950 HU and Exp > −856 HU (normal lung), Insp > −950 and Exp < −856 HU (SAD) and Insp < −950 and Exp −856HU (emphysema). The native CT data set (d) is overlaid by colour-coded volumetric representation of the densitometric categories that allow topographic description of pulmonary disease (green, normal lung; yellow, air trapping; red, emphysema). The present COPD case shows the coexistence of air trapping and emphysema, which are objectively apportioned by the b-phase densitometric quantification with evidence of airway predominant disease. COPD, chronic obstructive pulmonary disease; HU, Hounsfield unit; SAD, small airway disease.

Airway quantification

Quantitative analysis of the airway follows the preliminary processing step of bronchial tree segmentation and isolation of three-dimensional airway model^{89, 90} (Figure 4a–c). Airway is usually quantified according to its generation, and the generation-specific mean can be calculated throughout the entire lung,⁹¹ nonetheless, Gupta et al reported that a single bronchus might be representative.⁹² Direct characterization of airway is feasible for several bronchial generations,⁹³ yet, direct QCT metrics are widely accepted up to fifth generation.^{94, 95} Direct quantification of airway includes lumen diameter and area, wall thickness and area, relative area of wall (obtained by the ratio between area of the wall and total area of the airway at a given section) and density of the airway wall.⁹⁶ Smaller airway can be indirectly quantified by the regression line between the square root of the airway wall area and the internal perimeter of the airway: the so called Pi10 that reflects wall conspicuity in airway with internal perimeter of 10 mm (broadly 3 mm of internal diameter) (Figure 4d). The variability and potential bias of airway metrics are significant and include lung volume, age and transient inflammation.^{91, 97} For this reason, the clinical application of airway QCT is still to be validated.⁴ Pi10 is deemed the most consistent among airway metrics, albeit only when calculated on a minimum of 12 subsegmental bronchi.⁹⁸ Beyond size of bronchial components, also density was investigated and potential association with mast cell infiltration was found in asthmatic patients.⁹⁹

Figure 4. .

(a–d) Direct and indirect quantification of airway. Airway segmentation from trachea to intrapulmonary bronchi and stretched view of a selected airway in the right lower lobe (a); oblique axial reformatting of bronchial structure with direct automatic segmentation lumen diameter (LD: 4.6 mm), wall thickness (WT: 1.0 mm) and relative surface of wall compared to total airway surface (wall area, WA%: 52.1). Volumetric reconstruction of airway segmentation (c) for automatic direct measurement of all airways for indirect calculation of Pi10 by means of regression line derived from plotting of internal perimeter and square root of wall area of airways with internal perimeter <20 mm (d).

Expiratory scan is still an issue for direct quantification of airway. Preliminary experiences suggested that more severe asthma might be associated with increased airway stiffness at third and fourth generation bronchi.¹⁰⁰ However, more analysis is needed to have a clear picture on the consistency of this observation.

Vascular quantification

Vascular quantification is relatively complex task in quantitative imaging of the lung. Quantitative metrics of pulmonary vascular volume have been recently proposed and are utmost promising to fill the gaps between morphological descriptors and physiology. Diaz showed that the broncho-arterial ratio is increased in smokers because of a relative reduction in vessel size.¹⁰¹ More quantitative studies showed characteristic vascular pruning of small pulmonary arteries¹⁰² and reduction of lung perfusion assessed by contrast-enhanced dual-energy spectral CT.¹⁰³ The latter was first shown to detect perfusional defects in pulmonary embolism, with increasing degree of enhancement from occlusive to non-occlusive clots.¹⁰⁴ The spectral imaging for quantification of small pulmonary vessels is now being approached by extremely preliminary experience for detection of lung susceptibility to vasodilators in healthy smokers with emphysema,¹⁰³ yet, its application still yields some degree of variability (e.g. cardiac ejection rate, physiological gradients and scanning conditions) and artefacts (e.g. beam hardening, cardiac motion) that prevent such technique in the routine of COPD.^{105, 106} The role of vascular disease in the complex pathogenesis of COPD needs to be further investigated from different perspectives, either as a cause or a consequence of lung tissue depletion, and in association with cardiovascular function. Along the cardiopulmonary functional cascade, left ventricle filling and systemic blood delivery progressively decrease according to emphysema extent and airflow obstruction, as it was shown in a large cohort of COPD patients.¹⁰⁷

Clinical application

The hardest effort towards standardization of imaging data set (acquisition and reconstruction) is mandatory for clinical application of quantitative imaging.^108–110 In particular, continuous change in protocol acquisition for dose reduction and related evolution in reconstruction algorithm should be accounted.^{111, 112}

The clinical application of quantitative imaging looks now closer than ever, in particular in tertiary centres where multidisciplinary teams optimally merge information from physiology, symptoms, imaging and therapeutic options.

Treatment of emphysema

Lung volume reduction (LVR) is a therapeutic option to improve pulmonary mechanics in severe emphysema. Techniques of LVR include endobronchial valves (EBV), endobronchial coils and surgery (LVRS). Regional quantification of emphysema by CT is paramount for the planning of LVR.^{70, 113} Emphysema distribution pattern and fissures integrity are pivotal for prediction of treatment efficacy.

EBVs can be used in both upper and lower predominant emphysema and are the less invasive technique, yet their selection criteria are extremely strict: fissure integrity >90% for prediction of collateral ventilation (major contraindication to EBV), heterogeneity of emphysema distribution between lobes (>15% difference in emphysema extent between ipsilateral lobes), emphysema >40% in the target lobe¹¹⁴ (Figure 2). Furthermore, CT is extremely useful also for measurement of bronchial lumen and tailoring of EBV size. In the follow-up, CT can quantify the lobar volumetric reduction in treated lobes and the relative expansion of healthier lobes.

If EBV requirements are not met, endobronchial coils or LVRS can be considered. LVRS should be proposed only for upper lobe predominant emphysema. Both techniques are substantially irreversible, hence, they are preceded by a CT-based quantification of residual volume for estimation of post-procedural pulmonary function.¹¹⁵ For this purpose, the integration of lung damage by CT quantification and perfusion scintigraphy provides the best prediction of clinical outcome.¹¹⁶

COPD phenotypization

The relative contribution of airway and parenchymal disease varies substantially in COPD, and determines prognosis and therapeutic response. PFTs have limitation in differentiating between phenotypes,⁶⁸ thus, patients with the same GOLD stage may present substantial clinical differences. Conversely, the literature on QCT increasingly supports the substantial amount of functional information that can be extracted from CT for phenotypization of COPD, even in case of mild to moderate disease.^{117, 118} Integration of visual and quantitative CT assessment permits categorization of COPD into emphysema predominant subtypes (proposed five different patterns) and airway-predominant subtypes (proposed two patterns).¹¹⁹ Most subjects with emphysema have significant airway disease, conversely, a proportion of COPD subjects have minimal emphysematous lung (<6% of lung with density < −950 HU on inspiratory CT) and predominant airway disease.¹¹⁹ PRM for objective estimation of normal lung, air trapping and emphysema appears to be the most promising QCT tool for clinical phenotypization of COPD, with potential application in longitudinal follow-up (Figure 3).^{120, 121}

Bronchiolitis obliterans syndrome in transplanted patients

Bronchiolitis obliterans syndrome is seen after lung or bone marrow transplantation.^{72, 122} It is diagnosed by spirometry, though there is lack of clinical tools to identify the degree of small airway obstruction. Quantification of functional small airways disease by PRM was shown to offer prognostic stratification in transplant recipients with spirometric decline.^{72, 122} In particular, PRM-quantified air trapping >30% after lung transplant could outrank subjects with shorter survival among patients with decline in lung function.⁷²

Application in scientific investigation and clinical trials

Lung imaging is now being employed to provide quantitative assessment of morphology and function in scientific investigation. The utmost ability of imaging for non-invasive quantitative in vivo assessment of obstructive lung diseases increasingly provides imaging biomarkers as outcome measures in clinical trials.

Large longitudinal prospective trials recruited thousands of patients with and without COPD to undertake deep analysis about the clinical, functional, imaging and genetic framework of this syndrome.^123–125 These powerful trials represent the cornerstone of QCT investigation in COPD and provide massive data for its translation to clinical practice.^{4, 126,127} Major reports from these studies include regional disease progression,¹²¹ QCT metrics associated with COPD exacerbations,^{127, 128} quantification of cardiovascular disease in COPD,¹²⁹ QCT stratification of smokers without COPD,^{86, 130} variability of parenchymal QCT metrics according to smoking status⁸⁰ and standardization of CT protocol for multicentre application of QCT phenotypes.¹⁰⁸

Emphysema

Emphysema is irreversible disease by definition, and in some cases it is relatively fast progressive. It is the case of α−1 antitrypsin deficiency where parenchymal tissue is actively disrupted as a consequence of reduced protease inhibition. QCT of lung parenchyma can be used to quantify progression of emphysematous destruction. Stoeckley used the P15 to test the effect of intravenous α−1 antitrypsin against placebo and could quantify a relative reduction of emphysema progression in the pharmaceutical arm.¹³¹

Asthma severity: proximal and distal airway

The airway remodelling seen in asthmatic patients is a morphological feature that yields substantial promise of imaging biomarkers for personalized asthma care.^{132, 133} Airway remodelling is particularly targeted on QCT, notably, there is association between epithelial thickness and central airway quantification by CT.⁹⁴ This association led to definition of CT-based clinical clusters of asthma, with different clinical and therapeutic features. The imaging-based clustering of asthmatic patients was first proposed by Gupta, who found heterogeneity of response to bronchodilator in patients with different imaging cluster⁹² (Figure 5). Recently, further description of asthma clusters has been obtained including topographic metrics of air trapping, with substantial clinical relevance compared to asthma phenotypes based on the sole clinical characterization or sputum cell count.⁸⁸ Noteworthy, Choi reported that the four imaging-based clusters show different response to high-dose inhaled corticosteroids.⁸⁸

Figure 5. .

(a–c) Asthma phenotypes according to QCT cluster. Volumetric model of segmented airway and specific quantitative analysis of the right upper lobe apical segmental bronchus (RB1) on cross-section (insets). The airway metrics were normalized to the body surface area for definition of QCT clusters of asthma according to wall volume (WV) and lumen volume (LV). The three clusters are defined as follows: (a) cluster 1 with increased WV and LV, decreased percentage WV and severe air trapping; (b) cluster 2 with minor central airway remodelling, moderate air trapping and low response to bronchodilator; (c) cluster 3 reduced WV and LV, increased WV percentage and severe air trapping on CT. Figure reproduced under a Creative Commons license (CC BY) from Gupta S. et al, J Allergy Clin Immunol. 2014 Mar;133³ :729–38.e18. https://doi.org/10.1016/j.jaci.2013.09.039. LV, lumen volume; QCT, quantitative CT; WV, wall volume.

Longitudinal quantification of air trapping extent by CT has been used as biomarker to assess response to treatment,^134–136 furthermore, it was associated with vascular conspicuity in proximal airway wall.⁹⁴

This evidence and its logarithmic technical development foster readily available quantitative metrics to assess personalized asthma care in clinical trials and potential translation to clinical use of imaging-based asthma phenotypes.

Interstitial lung disease

The variability in clinical evaluation of ILD is a reason for automation, CAD, and quantitative image analysis.¹³⁷ Several studies showed that QCT is an objective analysis that may overcome the issue of the interobserver variability and could provide more consistent prognostic indexes.^138–142 Furthermore, QCT has the potential ability to identify CT features that are not visually recognizable and to objectively monitor the disease progression on serial CT scans.

State of the art

There are several quantitative CT systems of varying degrees of sophistication for the assessment of ILD. As opposed to pulmonary emphysema, ILD patterns are quite heterogeneous in morphologic characteristics and lack a standard density threshold that can dichotomize the visualized lung tissue into normal and diseased.¹⁴³ Nevertheless, the global histogram of density metrics of CT images—skewness, kurtosis and mean lung density—are helpful to estimate the ILD extent.^144–146 For instance, in pulmonary fibrosis, collagen deposition increases lung density, causing a rightward shift of the CT frequency histogram and reducing its peak (i.e. increasing skewness and kurtosis, respectively).¹⁴⁷ Furthermore, such metrics are sufficiently reproducible and not substantially affected by the reduction in radiation dose in subjects with ILD.^148–150

Lung volume variation due to different levels of inspiration may represent a major limitation of any density-based analysis of the lungs. Such a noise may be attenuated by evaluating the lung weight, which takes into account both lung volume and lung density.¹⁵¹

QCT was reported to enable objective tracking of the changes in lung weight and air-space inflation produced by a standard intervention, as in pulmonary alveolar proteinosis, suggesting lung weight could be a reliable metric to assess longitudinal change in ILDs (e.g. diffuse acute lung disorders).¹⁵² The density histogram parameters have also been used to quantify the extent of individual patterns of ILD (e.g. ground glass opacity, honeycombing, reticulation etc.).¹⁵³ This approach is not sufficient to achieve that goal, and more sophisticated textural analyses have been, therefore, implemented.² Parenchymal classification is applied to voxel volume unit (e.g. discrete volume that allows detailed characterization of local parenchymal features) using TA, computer vision-based image understanding of volumetric histogram signature mapping features and three-dimensional-morphology. Textural analysis is based on ROIs selected by trained observers in the lung, according to a set of specific patterns (normal, reticular, honeycombing etc.) (Figure 6). The histogram or textural features of each volumetric ROI are extracted, and a machine-learning algorithm is used to develop a predictive model for specific patterns.^{137,154–156} Given the well-known interobserver variability for the assessment of honeycombing, the development of an objective quantitative CT tool that can quantify honeycombing with prognostic value is of utmost importance. However, this kind of software analysis is limited by inbuilt subjectivity (e.g. owed to the expert observers pretraining), and other objective methods are currently being developed.^{2, 157}

Most textural-based software is still not commercially available on CT vendors’ diagnostic workstations and such software requires high-resolution images, preferably reconstructed with parameters that reduce image noise. Such multidimensional analysis demands considerable computational power that usually requires a dedicated workstation outside of the clinical radiology workflow.¹⁵⁴

There are still some important topics that need to be addressed in the future. First, it’s not fully clear if (and to what extent) CT technique optimization and standardization should be pursued for the quantitative analysis of ILD. This may have important implications for multicentre clinical trials that rely on accurate and reproducible quantitative analysis of CT images collected under varied conditions across multiple sites, scanners and time points.¹⁵⁷ Second, most QCT metrics of ILD severity are given as continuous data and are not, therefore, user-friendly for clinical practice. A staging system that defines ILD severity in categories (e.g. mild, moderate or severe) would be helpful for implementing QCT in clinical practice.¹³⁸

Application in scientific investigation and clinical trials

Most investigations have tested QCT tools in subjects with either idiopathic pulmonary fibrosis (IPF), or connective tissue disease (e.g. systemic sclerosis).^{141, 142,145,146,150,156,158} At present, automated image analysis of ILD is still confined to the research setting.

A large number of studies demonstrated that various QCT metrics correlated with several clinicofunctional indexes.^{139, 140,146,150,159,160} However, data on their prognostic value is of outmost importance. Best et al¹⁶¹ showed both kurtosis and visual scoring of the extent of fibrosing pattern as the only predictors of mortality in a retrospective study of 167 subjects with IPF recruited in a clinical trial. Recently, in a study of 46 subjects with IPF the histogram metrics correlated with PFT and were associated with transplant free survival similarly to the visual scoring performed by two experts.¹⁵³ Likewise, the histogram metrics can discriminate between well-defined different mortality risk categories in subjects with systemic sclerosis-related ILD.¹⁴²

Jacob showed that baseline texture-based CT quantification of total disease extent or individual patterns were superior to visual scoring in increasing the accuracy of clinicofunctional models predictive of outcome in IPF.¹³⁸ Intriguingly, the authors showed that the pulmonary vessels volume (PVV), was the individual QCT metric more strongly associated with mortality. Several hypotheses have been suggested, though the pathophysiological mechanism is not yet understood and further validation is required. Furthermore, they subsequently demonstrated that the PVV was also an independent predictor of mortality across patients with various connective tissue diseases.¹⁶²

The visual scoring of serial CTs is not fully standardized and QCT analysis may be particularly attractive for objectively monitoring IPF.² Maldonado showed that short-term (3–15 months) changes in CT patterns as assessed by the software was predictive of survival.¹⁴¹ Likewise, two recent studies using another software, showed that automatic quantification of lung fibrosis at CT yields an index of severity that correlates with visual assessment and functional change in subjects with IPF.^{156, 158}

However, the assessment of the severity of traction bronchiectasis—a major determinant of prognosis as visually quantified in subjects with either IPF or connective tissue disease—is still not allowed by any QCT tool.

Conclusion

There is a large amount of data that support the potential of QCT in pulmonary medicine. The fast technological development of such tools already brought to their clinical application, especially for the assessment of lung nodule and its management standardization. Furthermore, the prognostic yield of QCT in diffuse lung diseases is challenging the traditional approach based on clinical and functional assessment. Notably, QCT analysis has its strength in detailed volumetric characterization of lung parenchyma, and thus the potential to apportion the contribution of single heterogeneous determinants of lung disease. This characteristic appears particularly useful for clinical trials and, potentially, for selection of personalized treatment.

The future of QCT is granted by the logarithmic technological development that suggests computerized medical systems will integrate automatic data analysis in clinical practice for multidisciplinary prognostication and management of patient with pulmonary disease.

Figure 6. .

(a–j) Longitudinal quantitative analysis of fibrotic interstitial lung disease. Automatic volumetric segmentation of parenchymal abnormalities in a patient with idiopathic pulmonary fibrosis at baseline (top row: a−e) and 1-year follow-up (bottom row: f−j). The colour-coded overlay on native high resolution computed tomography (HRCT) images shows the distribution of parenchymal abnormalities on axial, coronal and sagittal reconstruction at baseline (a−c) and 1 year (f−h). The data are also provided in a volumetric model that shows both lungs with colour-coded characterization of parenchyma volume. Furthermore, a synthetic 2D graph (the so-called Glyph) is built that provides comprehensive display of abnormal parenchyma and its distribution between lobes (baseline Glyph in e, 1-year Glyph in j). 2D, two-dimensional.