Longitudinal assessment of interstitial lung abnormalities on CT in patients with COPD using artificial intelligence-based segmentation: a prospective observational study.

BACKGROUND: Interstitial lung abnormalities (ILAs) on CT may affect the clinical outcomes in patients with chronic obstructive pulmonary disease (COPD), but their quantification remains unestablished. This study examined whether artificial intelligence (AI)-based segmentation could be applied to identify ILAs using two COPD cohorts. METHODS: ILAs were diagnosed visually based on the Fleischner Society definition. Using an AI-based method, ground-glass opacities, reticulations, and honeycombing were segmented, and their volumes were summed to obtain the percentage ratio of interstitial lung disease-associated volume to total lung volume (ILDvol%). The optimal ILDvol% threshold for ILA detection was determined in cross-sectional data of the discovery and validation cohorts. The 5-year longitudinal changes in ILDvol% were calculated in discovery cohort patients who underwent baseline and follow-up CT scans. RESULTS: ILAs were found in 32 (14%) and 15 (10%) patients with COPD in the discovery (n = 234) and validation (n = 153) cohorts, respectively. ILDvol% was higher in patients with ILAs than in those without ILA in both cohorts. The optimal ILDvol% threshold in the discovery cohort was 1.203%, and good sensitivity and specificity (93.3% and 76.3%) were confirmed in the validation cohort. 124 patients took follow-up CT scan during 5 ± 1 years. 8 out of 124 patients (7%) developed ILAs. In a multivariable model, an increase in ILDvol% was associated with ILA development after adjusting for age, sex, BMI, and smoking exposure. CONCLUSION: AI-based CT quantification of ILDvol% may be a reproducible method for identifying and monitoring ILAs in patients with COPD.

Quantification of airway structures by persistent homology.

We propose two types of novel morphological metrics for quantifying the geometry of tubular structures on computed tomography (CT) images. We apply our metrics to identify irregularities in the airway of patients with chronic obstructive pulmonary disease (COPD) and demonstrate that they provide complementary information to the conventional metrics used to assess COPD, such as the tissue density distribution in lung parenchyma and the wall area ratio of the segmented airway. The three-dimensional shape of the airway and its abstraction as a rooted tree with the root at the trachea carina are automatically extracted from a lung CT volume, and the two metrics are computed based on a mathematical tool called persistent homology; treeH0 quantifies the distribution of branch lengths to assess the complexity of the tree-like structure and radialH0 quantifies the irregularities in the luminal radius along the airway. We show our metrics are associated with clinical outcomes.

A reference equation for lung volume on computed tomography in Japanese middle-aged and elderly adults.

BACKGROUND: Effective use of lung volume data measured on computed tomography (CT) requires reference values for specific populations. This study examined whether an equation previously generated for multiple ethnic groups in the United States, including Asians predominantly composed of Chinese people, in the Multi-Ethnic Study of Atherosclerosis (MESA) could be used for Japanese people and, if necessary, to optimize this equation. Moreover, the equation was used to characterize patients with chronic obstructive pulmonary disease (COPD) and lung hyperexpansion. METHODS: This study included a lung cancer screening CT cohort of asymptomatic never smokers aged ≥40 years from two institutions (n = 364 and 419) to validate and optimize the MESA equation and a COPD cohort (n = 199) to test its applicability. RESULTS: In all asymptomatic never smokers, the variance explained by the predicted values (R2) based on the original MESA equation was 0.60. The original equation was optimized to minimize the root mean squared error (RMSE) by adjusting the scaling factor but not the age, sex, height, or body mass index terms of the equation. The RMSE changed from 714 ml in the original equation to 637 ml in the optimized equation. In the COPD cohort, lung hyperexpansion, defined based on the 95th percentile of the ratio of measured lung volume to predicted lung volume in never smokers (122 %), was observed in 60 (30 %) patients and was associated with centrilobular emphysema and air trapping on inspiratory/expiratory CT. CONCLUSIONS: The MESA equation was optimized for Japanese middle-aged and elderly adults.

Centrilobular Emphysema Is Associated with Pectoralis Muscle Reduction in Current Smokers without Airflow Limitation.

BACKGROUND: Physiological and prognostic associations of centrilobular emphysema (CLE) and paraseptal emphysema (PSE) in smokers with and without chronic obstructive pulmonary disease (COPD) have been increasingly recognized, but the associations with extrapulmonary abnormalities, such as muscle wasting, osteoporosis, and cardiovascular diseases, remain unestablished. OBJECTIVES: The aim of the study was to investigate whether CLE was associated with extrapulmonary abnormalities independent of concomitant PSE in smokers without airflow limitation. METHODS: This retrospective study consecutively enrolled current smokers without airflow limitation who underwent lung cancer screening with computed tomography and spirometry. CLE and PSE were visually identified based on the Fleischner Society classification system. Cross-sectional areas of pectoralis muscles (PM) and adjacent subcutaneous adipose tissue (SAT), bone mineral density (BMD), and coronary artery calcification (CAC) were evaluated. RESULTS: Of 310 current smokers without airflow limitation, 83 (26.8%) had CLE. The PSE prevalence was higher (67.5% vs. 23.3%), and PM area, SAT area, and BMD were lower in smokers with CLE than in those without (PM area (mean), 34.5 versus 38.6 cm2; SAT area (mean), 29.3 versus 36.8 cm2; BMD (mean), 158.3 versus 178.4 Hounsfield unit), while CAC presence did not differ. In multivariable models, CLE was associated with lower PM area but not with SAT area or BMD, after adjusting for PSE presence, demographics, and forced expiratory volume in 1 s. CONCLUSIONS: The observed association between CLE and lower PM area suggests that susceptibility to skeletal muscle loss could be high in smokers with CLE even without COPD.

Subtyping emphysematous COPD by respiratory volume change distributions on CT.

BACKGROUND: There is considerable heterogeneity among patients with emphysematous chronic obstructive pulmonary disease (COPD). We hypothesised that in addition to emphysema severity, ventilation distribution in emphysematous regions would be associated with clinical-physiological impairments in these patients. OBJECTIVE: To evaluate whether the discordance between respiratory volume change distributions (from expiration to inspiration) in emphysematous and non-emphysematous regions affects COPD outcomes using two cohorts. METHODS: Emphysema was quantified using a low attenuation volume percentage on inspiratory CT (iLAV%). Local respiratory volume changes were calculated using non-rigidly registered expiratory/inspiratory CT. The Ventilation Discordance Index (VDI) represented the log-transformed Wasserstein distance quantifying discordance between respiratory volume change distributions in emphysematous and non-emphysematous regions. RESULTS: Patients with COPD in the first cohort (n=221) were classified into minimal emphysema (iLAV% <10%; n=113) and established emphysema with high VDI and low VDI groups (n=46 and 62, respectively). Forced expiratory volume in 1 s (FEV1) was lower in the low VDI group than in the other groups, with no difference between the high VDI and minimal emphysema groups. Higher iLAV%, more severe airway disease and hyperventilated emphysematous regions in the upper-middle lobes were independently associated with lower VDI. The second cohort analyses (n=93) confirmed these findings and showed greater annual FEV1 decline and higher mortality in the low VDI group than in the high VDI group independent of iLAV% and airway disease on CT. CONCLUSION: Lower VDI is associated with severe airflow limitation and higher mortality independent of emphysema severity and airway morphological changes in patients with emphysematous COPD.

Physiological impacts of computed tomography airway dysanapsis, fractal dimension, and branch count in asymptomatic never smokers.

Dysanapsis, a mismatch between airway tree caliber and lung size, contributes to a large variation in lung function on spirometry in healthy subjects. However, it remains unclear whether other morphological features of the airway tree could be associated with the variation in lung function independent of dysanapsis. This study used lung cancer screening chest computed tomography (CT) and spirometry data from asymptomatic never smokers. Dysanapsis and the complexity of airway tree geometry were quantified on CT by measuring airway to lung ratio (ALR) and airway fractal dimension (AFD). Moreover, total airway count (TAC), ratio of airway luminal surface area to volume (SA/V), longitudinal tapering and irregularity of the radius of the internal lumen from the central to peripheral airways (Tapering index and Irregularity index) were quantified. In 431 asymptomatic never smokers without a history of lung diseases, lower ALR was associated with lower forced expiratory volume in 1 s (FEV1) and FEV1/forced vital capacity (FEV1/FVC). The associations of ALR with AFD and TAC (r = 0.41 and 0.13) were weaker than the association between TAC and AFD (r = 0.64). In multivariable models adjusted for age, sex, height, and mean lung density, lower AFD and TAC were associated with lower FEV1 and FEV1/FVC independent of ALR, whereas SA/V and Tapering index were not. These results suggest that the smaller airway tree relative to a given lung size and the lower complexity of airway tree shape, including lower branch count, are independently associated with lower lung function in healthy subjects.NEW & NOTEWORTHY This study showed that fractal dimension and total airway count of the airway tree on computed tomography are associated with lung function on spirometry independent of a smaller airway for a given lung size (dysanapsis) in asymptomatic never smokers without a history of lung diseases. In addition to dysanapsis, the morphometric complexity of the airway tree and the airway branch count may cause a substantial variation of lung function in these subjects.

RNaseH2A downregulation drives inflammatory gene expression via genomic DNA fragmentation in senescent and cancer cells.

Cellular senescence caused by oncogenic stimuli is associated with the development of various age-related pathologies through the senescence-associated secretory phenotype (SASP). SASP is mediated by the activation of cytoplasmic nucleic acid sensors. However, the molecular mechanism underlying the accumulation of nucleotide ligands in senescent cells is unclear. In this study, we revealed that the expression of RNaseH2A, which removes ribonucleoside monophosphates (rNMPs) from the genome, is regulated by E2F transcription factors, and it decreases during cellular senescence. Residual rNMPs cause genomic DNA fragmentation and aberrant activation of cytoplasmic nucleic acid sensors, thereby provoking subsequent SASP factor gene expression in senescent cells. In addition, RNaseH2A expression was significantly decreased in aged mouse tissues and cells from individuals with Werner syndrome. Furthermore, RNaseH2A degradation using the auxin-inducible degron system induced the accumulation of nucleotide ligands and induction of certain tumourigenic SASP-like factors, promoting the metastatic properties of colorectal cancer cells. Our results indicate that RNaseH2A downregulation provokes SASP through nucleotide ligand accumulation, which likely contributes to the pathological features of senescent, progeroid, and cancer cells.

Training of deep cross-modality conversion models with a small data set, and their application in megavoltage CT to kilovoltage CT conversion.

PURPOSE: In recent years, deep learning-based image processing has emerged as a valuable tool for medical imaging owing to its high performance. However, the quality of deep learning-based methods heavily relies on the amount of training data; the high cost of acquiring a large data set is a limitation to their utilization in medical fields. Herein, based on deep learning, we developed a computed tomography (CT) modality conversion method requiring only a few unsupervised images. METHODS: The proposed method is based on cycle-consistency generative adversarial network (CycleGAN) with several extensions tailored for CT images, which aims at preserving the structure in the processed images and reducing the amount of training data. This method was applied to realize the conversion of megavoltage computed tomography (MVCT) to kilovoltage computed tomography (kVCT) images. Training was conducted using several data sets acquired from patients with head and neck cancer. The size of the data sets ranged from 16 slices (two patients) to 2745 slices (137 patients) for MVCT and 2824 slices (98 patients) for kVCT. RESULTS: The required size of the training data was found to be as small as a few hundred slices. By statistical and visual evaluations, the quality improvement and structure preservation of the MVCT images converted by the proposed model were investigated. As a clinical benefit, it was observed by medical doctors that the converted images enhanced the precision of contouring. CONCLUSIONS: We developed an MVCT to kVCT conversion model based on deep learning, which can be trained using only a few hundred unpaired images. The stability of the model against changes in data size was demonstrated. This study promotes the reliable use of deep learning in clinical medicine by partially answering commonly asked questions, such as "Is our data sufficient?" and "How much data should we acquire?"

A homological approach to a mathematical definition of pulmonary fibrosis and emphysema on computed tomography.

Three-dimensional imaging is essential to evaluate local abnormalities and understand structure-function relationships in an organ. However, quantifiable and interpretable methods to localize abnormalities remain unestablished. Visual assessments are prone to bias, machine learning methods depend on training images, and the underlying decision principle is usually difficult to interpret. Here, we developed a homological approach to mathematically define emphysema and fibrosis in the lungs on computed tomography (CT). With the use of persistent homology, the density of homological features, including connected components, tunnels, and voids, was extracted from the volumetric CT scans of lung diseases. A pair of CT values at which each homological feature appeared (birth) and disappeared (death) was computed by sweeping the threshold levels from higher to lower CT values. Consequently, fibrosis and emphysema were defined as voxels with dense voids having a longer lifetime (birth-death difference) and voxels with dense connected components having a lower birth, respectively. In an independent dataset including subjects with idiopathic pulmonary fibrosis (IPF), chronic obstructive pulmonary disease (COPD), and combined pulmonary fibrosis and emphysema (CPFE), the proposed definition enabled accurate segmentation with comparable quality to deep learning in terms of Dice coefficients. Persistent homology-defined fibrosis was closely associated with physiological abnormalities such as impaired diffusion capacity and long-term mortality in subjects with IPF and CPFE, and persistent homology-defined emphysema was associated with impaired diffusion capacity in subjects with COPD. The present persistent homology-based evaluation of structural abnormalities could help explore the clinical and physiological impacts of structural changes and morphological mechanisms of disease progression.NEW & NOTEWORTHY This study proposes a homological approach to mathematically define a three-dimensional texture feature of emphysema and fibrosis on chest computed tomography using persistent homology. The proposed definition enabled accurate segmentation with comparable quality to deep learning while offering higher interpretability than deep learning-based methods.

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

Chest computed tomography (CT) is used to screen for lung cancer and evaluate pulmonary and extra-pulmonary abnormalities such as emphysema and coronary artery calcification, particularly in smokers. In real-world practice, lung abnormalities are visually assessed using high-contrast thin-slice images which are generated from raw scan data using sharp reconstruction kernels with the sacrifice of increased image noise. In contrast, accurate CT quantification requires low-contrast thin-slice images with low noise, which are generated using soft reconstruction kernels. However, only sharp-kernel thin-slice images are archived in many medical facilities due to limited data storage space. This study aimed to establish deep neural network (DNN) models to convert sharp-kernel images to soft-kernel-like images with a final goal to reuse historical chest CT images for robust quantitative measurements, particularly in completed previous longitudinal studies. By using pairs of sharp-kernel (input) and soft-kernel (ground-truth) images from 30 patients with chronic obstructive pulmonary disease (COPD), DNN models were trained. Then, the accuracy of kernel conversion based on the established DNN models was evaluated using CT from independent 30 smokers with and without COPD. Consequently, differences in CT values between new images converted from sharp-kernel images using the established DNN models and ground-truth soft-kernel images were comparable with the inter-scans variability derived from repeated phantom scans (6 times), showing that the conversion error was the same level as the measurement error of the CT device. Moreover, the Dice coefficients to quantify the similarity between low attenuation voxels on given images and the ground-truth soft-kernel images were significantly higher on the DNN-converted images than the Gaussian-filtered, median-filtered, and sharp-kernel images (p < 0.001). There were good agreements in quantitative measurements of emphysema, intramuscular adipose tissue, and coronary artery calcification between the converted and the ground-truth soft-kernel images. These findings demonstrate the validity of the new DNN model for kernel conversion and the clinical applicability of soft-kernel-like images converted from archived sharp-kernel images in previous clinical studies. The presented method to evaluate the validity of the established DNN model using repeated scans of phantom could be applied to various deep learning-based image conversions for robust quantitative evaluation.

[Improvement in Image Quality of CBCT during Treatment by Cycle Generative Adversarial Network].

PURPOSE: Volumetric modulated arc therapy (VMAT) can acquire projection images during rotational irradiation, and cone-beam computed tomography (CBCT) images during VMAT delivery can be reconstructed. The poor quality of CBCT images prevents accurate recognition of organ position during the treatment. The purpose of this study was to improve the image quality of CBCT during the treatment by cycle generative adversarial network (CycleGAN). METHOD: Twenty patients with clinically localized prostate cancer were treated with VMAT, and projection images for intra-treatment CBCT (iCBCT) were acquired. Synthesis of PCT (SynPCT) with improved image quality by CycleGAN requires only unpaired and unaligned iCBCT and planning CT (PCT) images for training. We performed visual and quantitative evaluation to compare iCBCT, SynPCT and PCT deformable image registration (DIR) to confirm the clinical usefulness. RESULT: We demonstrated suitable CycleGAN networks and hyperparameters for SynPCT. The image quality of SynPCT improved visually and quantitatively while preserving anatomical structures of the original iCBCT. The undesirable deformation of PCT was reduced when SynPCT was used as its reference instead of iCBCT. CONCLUSION: We have performed image synthesis with preservation of organ position by CycleGAN for iCBCT and confirmed the clinical usefulness.

Assessment of dysplasia in bone marrow smear with convolutional neural network.

In this study, we developed the world's first artificial intelligence (AI) system that assesses the dysplasia of blood cells on bone marrow smears and presents the result of AI prediction for one of the most representative dysplasia-decreased granules (DG). We photographed field images from the bone marrow smears from patients with myelodysplastic syndrome (MDS) or non-MDS diseases and cropped each cell using an originally developed cell detector. Two morphologists labelled each cell. The degree of dysplasia was evaluated on a four-point scale: 0-3 (e.g., neutrophil with severely decreased granules were labelled DG3). We then constructed the classifier from the dataset of labelled images. The detector and classifier were based on a deep neural network pre-trained with natural images. We obtained 1797 labelled images, and the morphologists determined 134 DGs (DG1: 46, DG2: 77, DG3: 11). Subsequently, we performed a five-fold cross-validation to evaluate the performance of the classifier. For DG1-3 labelled by morphologists, the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy were 91.0%, 97.7%, 76.3%, 99.3%, and 97.2%, respectively. When DG1 was excluded in the process, the sensitivity, specificity, PPV, NPV, and accuracy were 85.2%, 98.9%, 80.6%, and 99.2% and 98.2%, respectively.

Visual enhancement of Cone-beam CT by use of CycleGAN.

PURPOSE: Cone-beam computed tomography (CBCT) offers advantages over conventional fan-beam CT in that it requires a shorter time and less exposure to obtain images. However, CBCT images suffer from low soft-tissue contrast, noise, and artifacts compared to conventional fan-beam CT images. Therefore, it is essential to improve the image quality of CBCT. METHODS: In this paper, we propose a synthetic approach to translate CBCT images with deep neural networks. Our method requires only unpaired and unaligned CBCT images and planning fan-beam CT (PlanCT) images for training. The CBCT images and PlanCT images may be obtained from other patients as long as they are acquired with the same scanner settings. Once trained, three-dimensionally reconstructed CBCT images can be directly translated into high-quality PlanCT-like images. RESULTS: We demonstrate the effectiveness of our method with images obtained from 20 prostate patients, and provide a statistical and visual comparison. The image quality of the translated images shows substantial improvement in voxel values, spatial uniformity, and artifact suppression compared to those of the original CBCT. The anatomical structures of the original CBCT images were also well preserved in the translated images. CONCLUSIONS: Our method produces visually PlanCT-like images from CBCT images while preserving anatomical structures.

Overview of image-to-image translation by use of deep neural networks: denoising, super-resolution, modality conversion, and reconstruction in medical imaging.

Since the advent of deep convolutional neural networks (DNNs), computer vision has seen an extremely rapid progress that has led to huge advances in medical imaging. Every year, many new methods are reported at conferences such as the International Conference on Medical Image Computing and Computer-Assisted Intervention and Machine Learning for Medical Image Reconstruction, or published online at the preprint server arXiv. There is a plethora of surveys on applications of neural networks in medical imaging (see [1] for a relatively recent comprehensive survey). This article does not aim to cover all aspects of the field, but focuses on a particular topic, image-to-image translation. Although the topic may not sound familiar, it turns out that many seemingly irrelevant applications can be understood as instances of image-to-image translation. Such applications include (1) noise reduction, (2) super-resolution, (3) image synthesis, and (4) reconstruction. The same underlying principles and algorithms work for various tasks. Our aim is to introduce some of the key ideas on this topic from a uniform viewpoint. We introduce core ideas and jargon that are specific to image processing by use of DNNs. Having an intuitive grasp of the core ideas of applications of neural networks in medical imaging and a knowledge of technical terms would be of great help to the reader for understanding the existing and future applications. Most of the recent applications which build on image-to-image translation are based on one of two fundamental architectures, called pix2pix and CycleGAN, depending on whether the available training data are paired or unpaired (see Sect. 1.3). We provide codes ([2, 3]) which implement these two architectures with various enhancements. Our codes are available online with use of the very permissive MIT license. We provide a hands-on tutorial for training a model for denoising based on our codes (see Sect. 6). We hope that this article, together with the codes, will provide both an overview and the details of the key algorithms and that it will serve as a basis for the development of new applications.

A circuit-preserving mapping from multilevel to Boolean dynamics.

Many discrete models of biological networks rely exclusively on Boolean variables and many tools and theorems are available for analysis of strictly Boolean models. However, multilevel variables are often required to account for threshold effects, in which knowledge of the Boolean case does not generalise straightforwardly. This motivated the development of conversion methods for multilevel to Boolean models. In particular, Van Ham's method has been shown to yield a one-to-one, neighbour and regulation preserving dynamics, making it the de facto standard approach to the problem. However, Van Ham's method has several drawbacks: most notably, it introduces vast regions of "non-admissible" states that have no counterpart in the multilevel, original model. This raises special difficulties for the analysis of interaction between variables and circuit functionality, which is believed to be central to the understanding of dynamic properties of logical models. Here, we propose a new multilevel to Boolean conversion method, with software implementation. Contrary to Van Ham's, our method doesn't yield a one-to-one transposition of multilevel trajectories; however, it maps each and every Boolean state to a specific multilevel state, thus getting rid of the non-admissible regions and, at the expense of (apparently) more complicated, "parallel" trajectories. One of the prominent features of our method is that it preserves dynamics and interaction of variables in a certain manner. As a demonstration of the usability of our method, we apply it to construct a new Boolean counter-example to the well-known conjecture that a local negative circuit is necessary to generate sustained oscillations. This result illustrates the general relevance of our method for the study of multilevel logical models.