Kernel Conversion for Robust Quantitative Measurements of Archived Chest Computed Tomography Using Deep Learning-Based Image-to-Image Translation

The Use of CT Densitometry for the Assessment of Emphysema in Clinical Trials: A Position Paper from the Fleischner Society

Emphysema's significant morbidity and mortality underscore the need for reliable outcome metrics in clinical trials. However, commonly accepted COPD outcome measures do not adequately capture emphysema severity or progression. Computed tomography (CT) metrics have been validated as accurate indicators of pathological emphysema and predictors of COPD progression, exacerbations, and mortality. This Position Paper reviews the evidence supporting CT densitometry as a biomarker for emphysema, establishes implementation standards, and highlights areas for future research. A systematic literature review addressed three key questions: whether CT densitometry can be used as a diagnostic biomarker of emphysema, whether CT densitometry can be used as prognostic biomarker, and whether longitudinal change in densitometry can be used as a disease progression monitoring biomarker. Emphysema metrics, such as the percentage of low attenuation areas (LAA-950), are validated, highly reproducible diagnostic and prognostic biomarkers. Volume-adjusted lung density is recommended for disease monitoring. Both metrics demonstrate a scan-rescan intra-class correlation coefficient of 0.99 with proper technique. The paper also discusses relevant CT physics, techniques, and sources of variation, including technical factors, physiological changes, and software analysis. Key recommendations for clinical trials include using standardized CT techniques, proper subject selection, and longitudinal evaluation with volume-adjusted lung density.

Establishing the effect of computed tomography reconstruction kernels on the measure of bone mineral density in opportunistic osteoporosis screening

Opportunistic computed tomography (CT) scans, which can assess relevant bones of interest, offer a potential solution for identifying osteoporotic individuals. However, it has been well documented that image protocol parameters, such as reconstruction kernel, impact the quantitative analysis of volumetric bone mineral density (vBMD) from CT scans. The purpose of this study was to investigate the impact that CT reconstruction kernels have on quantitative results for vBMD from clinical CT scans using phantom and internal calibration. 45 clinical CT scans were reconstructed using the standard kernel and seven alternative kernels: soft, chest, detail, edge, bone, bone plus and lung [GE HealthCare]. Two methods of image calibration, internal and phantom, were used to calibrate the scans. The total hip and fourth lumbar vertebra (L4) were extracted from the scans via deep learning segmentation. Integral vBMD was calculated based on both calibration techniques from CT scans reconstructed with the eight kernels. Linear regression and Bland-Altman analyses were used to determine the coefficient of determination [Formula: see text] and to quantify the agreement between the different kernels. Differences between the reconstruction kernels were determined using paired t tests, and mean differences from the standard were computed. Using internal calibration, the smoothest kernel (soft) yielded a mean difference of -0.95 mg/cc (-0.33%) compared to the reference standard at the L4 vertebra and 2.07 mg/cc (0.51%) at the left femur. The sharpest kernel (lung) yielded a mean difference of 25.36 mg/cc (9.63%) at the L4 vertebra and -25.10 mg/cc (-5.98%) at the left femur. Alternatively, using phantom calibration soft yielded higher mean differences than internal calibration at both locations, with mean differences of 1.21 mg/cc (0.42%) at the L4 vertebra and 2.53 mg/cc (0.65%) at the left femur. The most error-prone results stemmed from the use of the lung kernel, as this kernel displayed a mean difference of -21.90 mg/cc (-7.38%) and -17.24 mg/cc (-4.34%) at the L4 vertebra and femur, respectively. These results indicate when performing opportunistic CT analysis, errors due to interchanging smoothing kernels soft, chest and detail are negligible, but that interchanging between sharpening kernels (lung, bone, bone plus, edge) results in large errors that can significantly impact vBMD measures for osteoporosis screening and diagnosis.

A Physics-Informed Deep Neural Network for Harmonization of CT Images

OBJECTIVE

Computed Tomography (CT) quantification is affected by the variability in image acquisition and rendition. This paper aimed to reduce this variability by harmonizing the images utilizing physics-based deep neural networks (DNNs).

METHODS

An adversarial generative network was trained on virtual CT images acquired under various imaging conditions using a virtual imaging platform with 40 computational patient models. These models featured anthropomorphic lungs with different levels of pulmonary diseases, including nodules and emphysema. Imaging was conducted using a validated CT simulator at two dose levels and varying reconstruction kernels. The trained model was tested on an independent virtual test dataset and two clinical datasets.

RESULTS

On the virtual test set, the harmonizer improved the structural similarity index from 79.3 16.4% to 95.8 1.7%, normalized mean squared error from 16.7 9.7% to 9.2 1.7%, and peak signal-to-noise ratio from 27.7 3.7 dB to 32.2 1.6 dB. Moreover, the harmonized images yielded more precise quantification of emphysema-based imaging biomarkers for lung attenuation, LAA -950 from 5.6 8.7% to 0.23 0.16%, Perc 15 from 43.4 45.4 HU to 20.0 7.5 HU, and Lung Mass from 0.3 0.3 g to 0.1 0.2 g. In clinical data, the harmonizer reduced biomarker variability by an average of 70%. For lung nodules, harmonized images improved the detectability index by 6.5-fold and DNN-based precision by 6%.

CONCLUSION

The proposed harmonizer significantly enhances image quality and quantification accuracy in CT imaging.

SIGNIFICANCE

The study demonstrated the potential utility of image harmonization for consistent CT image quality and reliable quantification, which is crucial for clinical applications and patient management.

Spatial lung imaging in clinical and translational settings

For many severe lung diseases, non-invasive biomarkers from imaging could improve early detection of lung injury or disease onset, establish a diagnosis, or help follow-up disease progression and treatment strategies. Imaging of the thorax and lung is challenging due to its size, respiration movement, transferred cardiac pulsation, vast density range and gravitation sensitivity. However, there is extensive ongoing research in this fast-evolving field. Recent improvements in spatial imaging have allowed us to study the three-dimensional structure of the lung, providing both spatial architecture and transcriptomic information at single-cell resolution. This fast progression, however, comes with several challenges, including significant image file storage and network capacity issues, increased costs, data processing and analysis, the role of artificial intelligence and machine learning, and mechanisms to combine several modalities. In this review, we provide an overview of advances and current issues in the field of spatial lung imaging.

Lung CT harmonization of paired reconstruction kernel images using generative adversarial networks

BACKGROUND

The kernel used in CT image reconstruction is an important factor that determines the texture of the CT image. Consistency of reconstruction kernel choice is important for quantitative CT-based assessment as kernel differences can lead to substantial shifts in measurements unrelated to underlying anatomical structures.

PURPOSE

In this study, we investigate kernel harmonization in a multi-vendor low-dose CT lung cancer screening cohort and evaluate our approach's validity in quantitative CT-based assessments.

METHODS

Using the National Lung Screening Trial, we identified CT scan pairs of the same sessions with one reconstructed from a soft tissue kernel and one from a hard kernel. In total, 1000 pairs of five different paired kernel types (200 each) were identified. We adopt the pix2pix architecture to train models for kernel conversion. Each model was trained on 100 pairs and evaluated on 100 withheld pairs. A total of 10 models were implemented. We evaluated the efficacy of kernel conversion based on image similarity metrics including root mean squared error (RMSE), peak signal-to-noise ratio (PSNR), and structural similarity index measure (SSIM) as well as the capability of the models to reduce measurement shifts in quantitative emphysema and body composition measurements. Additionally, we study the reproducibility of standard radiomic features for all kernel pairs before and after harmonization.

RESULTS

Our approach effectively converts CT images from one kernel to another in all paired kernel types, as indicated by the reduction in RMSE (p < 0.05) and an increase in the PSNR (p < 0.05) and SSIM (p < 0.05) for both directions of conversion for all pair types. In addition, there is an increase in the agreement for percent emphysema, skeletal muscle area, and subcutaneous adipose tissue (SAT) area for both directions of conversion. Furthermore, radiomic features were reproducible when compared with the ground truth features.

CONCLUSIONS

Kernel conversion using deep learning reduces measurement variation in percent emphysema, muscle area, and SAT area.

Deep learning in computed tomography super resolution using multi-modality data training

BACKGROUND

One of the limitations in leveraging the potential of artificial intelligence in X-ray imaging is the limited availability of annotated training data. As X-ray and CT shares similar imaging physics, one could achieve cross-domain data sharing, so to generate labeled synthetic X-ray images from annotated CT volumes as digitally reconstructed radiographs (DRRs). To account for the lower resolution of CT and the CT-generated DRRs as compared to the real X-ray images, we propose the use of super-resolution (SR) techniques to enhance the CT resolution before DRR generation.

PURPOSE

As spatial resolution can be defined by the modulation transfer function kernel in CT physics, we propose to train a SR network using paired low-resolution (LR) and high-resolution (HR) images by varying the kernel's shape and cutoff frequency. This is different to previous deep learning-based SR techniques on RGB and medical images which focused on refining the sampling grid. Instead of generating LR images by bicubic interpolation, we aim to create realistic multi-detector CT (MDCT) like LR images from HR cone-beam CT (CBCT) scans.

METHODS

We propose and evaluate the use of a SR U-Net for the mapping between LR and HR CBCT image slices. We reconstructed paired LR and HR training volumes from the same CT scans with small in-plane sampling grid size of0.20 × 0.20 mm 2 $0.20 \times 0.20 \, {\rm mm}^2$ . We used the residual U-Net architecture to train two models. SRUNR e s K $^K_{Res}$ : trained with kernel-based LR images, and SRUNR e s I $^I_{Res}$ : trained with bicubic downsampled data as baseline. Both models are trained on one CBCT dataset (n = 13 391). The performance of both models was then evaluated on unseen kernel-based and interpolation-based LR CBCT images (n = 10 950), and also on MDCT images (n = 1392).

RESULTS

Five-fold cross validation and ablation study were performed to find the optimal hyperparameters. Both SRUNR e s K $^K_{Res}$ and SRUNR e s I $^I_{Res}$ models show significant improvements (p-value < $<$ 0.05) in mean absolute error (MAE), peak signal-to-noise ratio (PSNR) and structural similarity index measures (SSIMs) on unseen CBCT images. Also, the improvement percentages in MAE, PSNR, and SSIM by SRUNR e s K $^K_{Res}$ is larger than SRUNR e s I $^I_{Res}$ . For SRUNR e s K $^K_{Res}$ , MAE is reduced by 14%, and PSNR and SSIMs increased by 6 and 8%, respectively. To conclude, SRUNR e s K $^K_{Res}$ outperforms SRUNR e s I $^I_{Res}$ , which the former generates sharper images when tested with kernel-based LR CBCT images as well as cross-modality LR MDCT data.

CONCLUSIONS

Our proposed method showed better performance than the baseline interpolation approach on unseen LR CBCT. We showed that the frequency behavior of the used data is important for learning the SR features. Additionally, we showed cross-modality resolution improvements to LR MDCT images. Our approach is, therefore, a first and essential step in enabling realistic high spatial resolution CT-generated DRRs for deep learning training.

Deep Learning-Based Kernel Adaptation Enhances Quantification of Emphysema on Low-Dose Chest CT for Predicting Long-Term Mortality

OBJECTIVES

The aim of this study was to ascertain the predictive value of quantifying emphysema using low-dose computed tomography (LDCT) post deep learning-based kernel adaptation on long-term mortality.

MATERIALS AND METHODS

This retrospective study investigated LDCTs obtained from asymptomatic individuals aged 60 years or older during health checkups between February 2009 and December 2016. These LDCTs were reconstructed using a 1- or 1.25-mm slice thickness alongside high-frequency kernels. A deep learning algorithm, capable of generating CT images that resemble standard-dose and low-frequency kernel images, was applied to these LDCTs. To quantify emphysema, the lung volume percentage with an attenuation value less than or equal to -950 Hounsfield units (LAA-950) was gauged before and after kernel adaptation. Low-dose chest CTs with LAA-950 exceeding 6% were deemed emphysema-positive according to the Fleischner Society statement. Survival data were sourced from the National Registry Database at the close of 2021. The risk of nonaccidental death, excluding causes such as injury or poisoning, was explored according to the emphysema quantification results using multivariate Cox proportional hazards models.

RESULTS

The study comprised 5178 participants (mean age ± SD, 66 ± 3 years; 3110 males). The median LAA-950 (18.2% vs 2.6%) and the proportion of LDCTs with LAA-950 exceeding 6% (96.3% vs 39.3%) saw a significant decline after kernel adaptation. There was no association between emphysema quantification before kernel adaptation and the risk of nonaccidental death. Nevertheless, after kernel adaptation, higher LAA-950 (hazards ratio for 1% increase, 1.01; P = 0.045) and LAA-950 exceeding 6% (hazards ratio, 1.36; P = 0.008) emerged as independent predictors of nonaccidental death, upon adjusting for age, sex, and smoking status.

CONCLUSIONS

The application of deep learning for kernel adaptation proves instrumental in quantifying pulmonary emphysema on LDCTs, establishing itself as a potential predictive tool for long-term nonaccidental mortality in asymptomatic individuals.

Centrilobular emphysema and airway dysanapsis: factors associated with low respiratory function in younger smokers

Background

Low respiratory function in young adulthood is one of the important factors in the trajectory leading to the future development of COPD, but its morphological characteristics are not well characterised.

Methods

We retrospectively enrolled 172 subjects aged 40-49 years with ≥10 pack-years smoking history who underwent lung cancer screening by computed tomography (CT) and spirometry at two Japanese hospitals. Emphysema was visually assessed according to the Fleischner Society guidelines and classified into two types: centrilobular emphysema (CLE) and paraseptal emphysema (PSE). Airway dysanapsis was assessed with the airway/lung ratio (ALR), which was calculated by the geometric mean of the lumen diameters of the 14 branching segments divided by the cube root of total lung volume on a CT scan.

Results

Among the subjects, CLE and PSE were observed in 20.9% and 30.8%, respectively. The mean ALR was 0.04 and did not differ between those with and without each type of emphysema. Multivariable regression analysis models adjusted for age, sex, body mass index and smoking status indicated that CLE and a low ALR were independently associated with lower forced expiratory volume in 1 s (FEV1)/forced vital capacity (estimate -1.64 (95% CI -2.68- -0.60) and 6.73 (95% CI 4.24-9.24), respectively) and FEV1 % pred (estimate -2.81 (95% CI -5.10- -0.52) and 10.9 (95% CI 5.36-16.4), respectively).

Conclusions

CLE and airway dysanapsis on CT were independently associated with low respiratory function in younger smokers.

Inter-vendor harmonization of CT reconstruction kernels using unpaired image translation

The reconstruction kernel in computed tomography (CT) generation determines the texture of the image. Consistency in reconstruction kernels is important as the underlying CT texture can impact measurements during quantitative image analysis. Harmonization (i.e., kernel conversion) minimizes differences in measurements due to inconsistent reconstruction kernels. Existing methods investigate harmonization of CT scans in single or multiple manufacturers. However, these methods require paired scans of hard and soft reconstruction kernels that are spatially and anatomically aligned. Additionally, a large number of models need to be trained across different kernel pairs within manufacturers. In this study, we adopt an unpaired image translation approach to investigate harmonization between and across reconstruction kernels from different manufacturers by constructing a multipath cycle generative adversarial network (GAN). We use hard and soft reconstruction kernels from the Siemens and GE vendors from the National Lung Screening Trial dataset. We use 50 scans from each reconstruction kernel and train a multipath cycle GAN. To evaluate the effect of harmonization on the reconstruction kernels, we harmonize 50 scans each from Siemens hard kernel, GE soft kernel and GE hard kernel to a reference Siemens soft kernel (B30f) and evaluate percent emphysema. We fit a linear model by considering the age, smoking status, sex and vendor and perform an analysis of variance (ANOVA) on the emphysema scores. Our approach minimizes differences in emphysema measurement and highlights the impact of age, sex, smoking status and vendor on emphysema quantification.

Harmonizing CT Images via Physics-based Deep Neural Networks

The rendition of medical images influences the accuracy and precision of quantifications. Image variations or biases make measuring imaging biomarkers challenging. The objective of this paper is to reduce the variability of computed tomography (CT) quantifications for radiomics and biomarkers using physics-based deep neural networks (DNNs). With the proposed framework, it is possible to harmonize the different renditions of a single CT scan (with variations in reconstruction kernel and dose) into an image that is in close agreement with the ground truth. To this end, a generative adversarial network (GAN) model was developed where the generator is informed by the scanner's modulation transfer function (MTF). To train the network, a virtual imaging trial (VIT) platform was used to acquire CT images, from a set of forty computational models (XCAT) serving as the patient model. Phantoms with varying levels of pulmonary disease, such as lung nodules and emphysema, were used. We scanned the patient models with a validated CT simulator (DukeSim) modeling a commercial CT scanner at 20 and 100 mAs dose levels and then reconstructed the images by twelve kernels representing smooth to sharp kernels. An evaluation of the harmonized virtual images was conducted in four different ways: 1) visual quality of the images, 2) bias and variation in density-based biomarkers, 3) bias and variation in morphological-based biomarkers, and 4) Noise Power Spectrum (NPS) and lung histogram. The trained model harmonized the test set images with a structural similarity index of 0.95±0.1, a normalized mean squared error of 10.2±1.5%, and a peak signal-to-noise ratio of 31.8±1.5 dB. Moreover, emphysema-based imaging biomarkers of LAA-950 (-1.5±1.8), Perc15 (13.65±9.3), and Lung mass (0.1±0.3) had more precise quantifications.

Fully automatic coronary calcium scoring in non-ECG-gated low-dose chest CT: comparison with ECG-gated cardiac CT

OBJECTIVES

To validate an artificial intelligence (AI)-based fully automatic coronary artery calcium (CAC) scoring system on non-electrocardiogram (ECG)-gated low-dose chest computed tomography (LDCT) using multi-institutional datasets with manual CAC scoring as the reference standard.

METHODS

This retrospective study included 452 subjects from three academic institutions, who underwent both ECG-gated calcium scoring computed tomography (CSCT) and LDCT scans. For all CSCT and LDCT scans, automatic CAC scoring (CAC_auto) was performed using AI-based software, and manual CAC scoring (CAC_man) was set as the reference standard. The reliability and agreement of CAC_auto was evaluated and compared with that of CAC_man using intraclass correlation coefficients (ICCs) and Bland-Altman plots. The reliability between CAC_auto and CAC_man for CAC severity categories was analyzed using weighted kappa (κ) statistics.

RESULTS

CAC_auto on CSCT and LDCT yielded a high ICC (0.998, 95% confidence interval (CI) 0.998-0.999 and 0.989, 95% CI 0.987-0.991, respectively) and a mean difference with 95% limits of agreement of 1.3 ± 37.1 and 0.8 ± 75.7, respectively. CAC_auto achieved excellent reliability for CAC severity (κ = 0.918-0.972) on CSCT and good to excellent but heterogenous reliability among datasets (κ = 0.748-0.924) on LDCT.

CONCLUSIONS

The application of an AI-based automatic CAC scoring software to LDCT shows good to excellent reliability in CAC score and CAC severity categorization in multi-institutional datasets; however, the reliability varies among institutions.

KEY POINTS

• AI-based automatic CAC scoring on LDCT shows excellent reliability with manual CAC scoring in multi-institutional datasets. • The reliability for CAC score-based severity categorization varies among datasets. • Automatic scoring for LDCT shows a higher false-positive rate than automatic scoring for CSCT, and most common causes of a false-positive are image noise and artifacts for both CSCT and LDCT.

The Potential Role of Artificial Intelligence in Lung Cancer Screening Using Low-Dose Computed Tomography

Two large randomized controlled trials of low-dose CT (LDCT)-based lung cancer screening (LCS) in high-risk smoker populations have shown a reduction in the number of lung cancer deaths in the screening group compared to a control group. Even if various countries are currently considering the implementation of LCS programs, recurring doubts and fears persist about the potentially high false positive rates, cost-effectiveness, and the availability of radiologists for scan interpretation. Artificial intelligence (AI) can potentially increase the efficiency of LCS. The objective of this article is to review the performances of AI algorithms developed for different tasks that make up the interpretation of LCS CT scans, and to estimate how these AI algorithms may be used as a second reader. Despite the reduction in lung cancer mortality due to LCS with LDCT, many smokers die of comorbid smoking-related diseases. The identification of CT features associated with these comorbidities could increase the value of screening with minimal impact on LCS programs. Because these smoking-related conditions are not systematically assessed in current LCS programs, AI can identify individuals with evidence of previously undiagnosed cardiovascular disease, emphysema or osteoporosis and offer an opportunity for treatment and prevention.