Search

Background: Timely diagnosis of structural heart disease improves patient outcomes, yet many remain underdiagnosed. While population screening with echocardiography is impractical, ECG-based prediction models can help target high-risk patients. We developed a novel ECG-based machine learning approach to predict multiple structural heart conditions, hypothesizing that a composite model would yield higher prevalence and positive predictive values to facilitate meaningful recommendations for echocardiography. Methods: Using 2 232 130 ECGs linked to electronic health records and echocardiography reports from 484 765 adults between 1984 to 2021, we trained machine learning models to predict the presence or absence of any of 7 echocardiography-confirmed diseases within 1 year. This composite label included the following: moderate or severe valvular disease (aortic/mitral stenosis or regurgitation, tricuspid regurgitation), reduced ejection fraction 15 mm. We tested various combinations of input features (demographics, laboratory values, structured ECG data, ECG traces) and evaluated model performance using 5-fold cross-validation, multisite validation trained on 1 site and tested on 10 independent sites, and simulated retrospective deployment trained on pre-2010 data and deployed in 2010. Results: Our composite rECHOmmend model used age, sex, and ECG traces and had a 0.91 area under the receiver operating characteristic curve and a 42% positive predictive value at 90% sensitivity, with a composite label prevalence of 17.9%. Individual disease models had area under the receiver operating characteristic curves from 0.86 to 0.93 and lower positive predictive values from 1% to 31%. Area under the receiver operating characteristic curves for models using different input features ranged from 0.80 to 0.93, increasing with additional features. Multisite validation showed similar results to cross-validation, with an aggregate area under the receiver operating characteristic curve of 0.91 across our independent test set of 10 clinical sites after training on a separate site. Our simulated retrospective deployment showed that for ECGs acquired in patients without preexisting structural heart disease in the year 2010, 11% were classified as high risk and 41% (4.5% of total patients) developed true echocardiography-confirmed disease within 1 year. Conclusions: An ECG-based machine learning model using a composite end point can identify a high-risk population for having undiagnosed, clinically significant structural heart disease while outperforming single-disease models and improving practical utility with higher positive predictive values. This approach can facilitate targeted screening with echocardiography to improve underdiagnosis of structural heart disease.

Several large trials have employed age or clinical features to select patients for atrial fibrillation (AF) screening to reduce strokes. We hypothesized that a machine learning (ML) model trained to predict AF risk from 12‑lead electrocardiogram (ECG) would be more efficient than criteria based on clinical variables in indicating a population for AF screening to potentially prevent AF-related stroke.We retrospectively included all patients with clinical encounters in Geisinger without a prior history of AF. Incidence of AF within 1 year and AF-related strokes within 3 years of the encounter were identified. AF-related stroke was defined as a stroke where AF was diagnosed at the time of stroke or within a year after the stroke. The efficiency of five methods was evaluated for selecting a cohort for AF screening. The methods were selected from four clinical trials (mSToPS, GUARD-AF, SCREEN-AF and STROKESTOP) and the ECG-based ML model. We simulated patient selection for the five methods between the years 2011 and 2014 and evaluated outcomes for 1 year intervals between 2012 and 2015, resulting in a total of twenty 1-year periods. Patients were considered eligible if they met the criteria before the start of the given 1-year period or within that period. The primary outcomes were numbers needed to screen (NNS) for AF and AF-associated stroke.The clinical trial models indicated large proportions of the population with a prior ECG for AF screening (up to 31%), coinciding with NNS ranging from 14 to 18 for AF and 249-359 for AF-associated stroke. At comparable sensitivity, the ECG ML model indicated a modest number of patients for screening (14%) and had the highest efficiency in NNS for AF (7.3; up to 60% reduction) and AF-associated stroke (223; up to 38% reduction).An ECG-based ML risk prediction model is more efficient than contemporary AF-screening criteria based on age alone or age and clinical features at indicating a population for AF screening to potentially prevent AF-related strokes.

Background Heart failure is a prevalent, costly disease for which new value-based payment models demand optimized population management strategies. Objectives This study sought to generate a strategy for managing populations of patients with heart failure by leveraging large clinical datasets and machine learning. Methods Geisinger electronic health record data were used to train machine learning models to predict 1-year all-cause mortality in 26,971 patients with heart failure who underwent 276,819 clinical episodes. There were 26 clinical variables (demographics, laboratory test results, medications), 90 diagnostic codes, 41 electrocardiogram measurements and patterns, 44 echocardiographic measurements, and 8 evidence-based “care gaps”: flu vaccine, blood pressure of Results Machine learning models achieved areas under the receiver-operating characteristic curve (AUCs) of 0.74 to 0.77 in a split-by-year training/test scheme, with the nonlinear XGBoost model (AUC: 0.77) outperforming linear logistic regression (AUC: 0.74). Out of 13,238 currently living patients, 2,844 were predicted to die within a year, and closing all care gaps was predicted to save 231 of these lives. Prioritizing patients for intervention by using the predicted reduction in 1-year mortality risk outperformed all other priority rankings (e.g., random selection or Seattle Heart Failure risk score). Conclusions Machine learning can be used to priority-rank patients most likely to benefit from interventions to optimize evidence-based therapies. This approach may prove useful for optimizing heart failure population health management teams within value-based payment models.

Use of machine learning for automated annotation of heart structures from echocardiographic videos is an active research area, but understanding of comparative, generalizable performance among models is lacking. This study aimed to 1) assess the generalizability of five state-of-the-art machine learning-based echocardiography segmentation models within a large clinical dataset, and 2) test the hypothesis that a quality control (QC) method based on segmentation uncertainty can further improve segmentation results. Five models were applied to 47,431 echocardiography studies that were independent from any training samples. Chamber volume and mass from model segmentations were compared to clinically-reported values. The median absolute errors (MAE) in left ventricular (LV) volumes and ejection fraction exhibited by all five models were comparable to reported inter-observer errors (IOE). MAE for left atrial volume and LV mass were similarly favorable to respective IOE for models trained for those tasks. A single model consistently exhibited the lowest MAE in all five clinically-reported measures. We leveraged the 10-fold cross-validation training scheme of this best-performing model to quantify segmentation uncertainty for potential application as QC. We observed that filtering segmentations with high uncertainty improved segmentation results, leading to decreased volume/mass estimation errors. The addition of contour-convexity filters further improved QC efficiency. In conclusion, five previously published echocardiography segmentation models generalized to a large, independent clinical dataset—segmenting one or multiple cardiac structures with overall accuracy comparable to manual analyses—with variable performance. Convexity-reinforced uncertainty QC efficiently improved segmentation performance and may further facilitate the translation of such models.

Use of machine learning (ML) for automated annotation of heart structures from echocardiographic videos is an active research area, but understanding of comparative, generalizable performance among models is lacking. This study aimed to (1) assess the generalizability of five state-of-the-art ML-based echocardiography segmentation models within a large Geisinger clinical dataset, and (2) test the hypothesis that a quality control (QC) method based on segmentation uncertainty can further improve segmentation results. Five models were applied to 47,431 echocardiography studies that were independent from any training samples. Chamber volume and mass from model segmentations were compared to clinically-reported values. The median absolute errors (MAE) in left ventricular (LV) volumes and ejection fraction exhibited by all five models were comparable to reported inter-observer errors (IOE). MAE for left atrial volume and LV mass were similarly favorable to respective IOE for models trained for those tasks. A single model consistently exhibited the lowest MAE in all five clinically-reported measures. We leveraged the tenfold cross-validation training scheme of this best-performing model to quantify segmentation uncertainty. We observed that removing segmentations with high uncertainty from 14 to 71% studies reduced volume/mass MAE by 6-10%. The addition of convexity filters improved specificity, efficiently removing 10% studies with large MAE (16-40%). In conclusion, five previously published echocardiography segmentation models generalized to a large, independent clinical dataset-segmenting one or multiple cardiac structures with overall accuracy comparable to manual analyses-with variable performance. Convexity-reinforced uncertainty QC efficiently improved segmentation performance and may further facilitate the translation of such models.

Introduction: Timely diagnosis of structural heart disease improves patient outcomes, yet millions remain undiagnosed. ECG-based prediction models can help identify high-risk patients for targeted screening, but existing individual disease models often have low positive predictive values (PPV) and limited clinical utility. Hypothesis: An ECG-based composite model can predict one of multiple, actionable structural heart conditions and yield higher prevalence and PPVs than individual models. Methods: Using 2,141,366 ECGs linked to echocardiography and EHR records from 461,466 adults from 1984-2021, we trained machine learning models to predict any of 7 echocardiography-confirmed diseases within 1 year. This composite label included: moderate or severe valvular disease (aortic stenosis or regurgitation, mitral stenosis or regurgitation, tricuspid regurgitation), reduced ejection fraction of 15mm. We tested various combinations of inputs and evaluated model performance with 1) cross-validation and 2) a simulated retrospective deployment. We measured area under the receiver operating characteristic curve (AUROC) and precision-recall curve (AUPRC), and PPV at 90% sensitivity. Results: Our composite "rECHOmmend" model using age, sex and ECG traces had an AUROC of 0.91, AUPRC of 0.78, and PPV of 52% at 90% sensitivity and 23% disease prevalence. Individual disease models had similar AUROCs (0.88-0.93), but lower AUPRCs (0.07-0.71) and PPVs (2%-41%; Figure). Across inputs, model AUROCs ranged from 0.85 to 0.93. Our simulated deployment model classified 22% of at-risk patients in 2010 as high-risk, of whom 40% developed true, echo-confirmed disease within 1 year. Conclusions: An ECG-based machine learning model using a composite endpoint can predict undiagnosed structural heart disease, outperforming single disease models with higher PPVs to facilitate targeted screening with echocardiography.

Background: Several large trials have employed age or clinical features to select patients for atrial fibrillation (AF) screening to reduce strokes. We hypothesized that a deep neural network (DNN) model risk prediction based on ECG would be superior to age and clinical variables at selecting a population at high risk for AF and AF-related stroke. Methods: We retrospectively included all patients with an ECG at Geisinger without a prior history of AF. Incidence of AF and AF-related strokes were identified as outcomes within 1 and 3 years after the ECG, respectively. AF-related stroke was defined as a stroke where AF was diagnosed at the time of stroke or within a year after the stroke. We selected a high-risk cohort for AF screening based on five risk stratification methods - criteria from four clinical trials (mSToPS, STROKESTOP, GUARD-AF and SCREEN-AF) and the DNN model at the qualifying ECG. We simulated patient selection and evaluated outcomes for twenty 1-year periods between 2010-2014 centered around the ECG encounter. For the clinical trials, the patients were considered eligible if they met the criteria before or within the period unless they satisfied exclusion criteria at the time of ECG. Results: The DNN model achieved optimal sensitivity (65%), PPV (10%), NNS for AF (10) within this population compared with all other risk models with a NNS for AF-related stroke of 160. Total screening number, sensitivity, positive predictive value (PPV) and number needed to screen (NNS) to capture AF and AF-related stroke are summarized in Table 1. The number of additional screens for the DNN model was slightly higher for two of the other models (SCREEN-AF and STROKESTOP) but lower than the other two (mSToPS and GUARD-AF). Conclusions: A DNN ECG-based risk prediction model is superior to contemporary AF-screening criteria based on age alone or age and clinical features in selecting a population for additional screening due to high risk for future AF and potential AF-related strokes.

Introduction: Initiation of QTc-prolonging medications may lead to the rare but potentially catastrophic event, torsades de pointes (TDP). At present, no adequate, generalizable tools exist to predict drug-induced long QTc (LQT); machine learning from ECG data is a promising approach. Hypothesis: Prediction of drug-induced LQT using an ECG-based machine learning model is feasible, and outperforms a model trained on baseline QTc, age, and sex alone. Methods: We identified baseline 12-lead ECGs with QTc values < 500 ms for patients who had not received any known, conditional, or possible QTc prolonging medication per CredibleMeds at the time of ECG or within the past 90 days. We matched these with ECGs from the same patients while they were taking at least one CredibleMeds drug (“on-drug” ECGs). Using 5-fold cross-validation, we trained and tested two machine learning models using the baseline ECGs of the 92,848 resulting pairs to predict drug-induced LQT (≥500 ms) in the on-drug ECGs: a deep neural network using ECG voltage data, and a gradient-boosted tree using the baseline QTc. Age and sex were also inputs to both models. Results: On-drug LQT prevalence was 16%. The ECG model demonstrated superior performance in predicting on-drug LQT (area under the receiver operating characteristic curve (AUC) = 0.756) compared to the QTc model (0.710). At a potential operating point (Figure), the ECG model had 89% sensitivity and 95% negative predictive value. Even in the subset of patients with baseline QTc < 470/480 ms (male/female; post-drug LQT prevalence = 14%), the ECG model demonstrated good performance (AUC = 0.736). Conclusions: An ECG-based machine learning model can stratify patients by risk of developing drug-induced LQT better than a model using baseline QTc alone. This model may have clinical value to identify high-risk drug starts that would benefit from closer monitoring and others who are at low risk of drug-induced LQT.

BackgroundEarly diagnosis of structural heart disease improves patient outcomes, yet many remain underdiagnosed. While population screening with echocardiography is impractical, electrocardiogram (ECG)-based prediction models can help target high-risk patients. We developed a novel ECG-based machine learning approach to predict multiple structural heart conditions, hypothesizing that a composite model would yield higher prevalence and positive predictive values (PPVs) to facilitate meaningful recommendations for echocardiography.MethodsUsing 2,232,130 ECGs linked to electronic health records and echocardiography reports from 484,765 adults between 1984-2021, we trained machine learning models to predict the presence of any of seven echocardiography-confirmed diseases within one year. This composite label included: moderate or severe valvular disease (aortic/mitral stenosis or regurgitation, tricuspid regurgitation), reduced ejection fraction 15mm. We tested various combinations of input features (demographics, labs, structured ECG data, ECG traces) and evaluated model performance using 5-fold cross-validation, multi-site validation trained on one clinical site and tested on 11 other independent sites, and simulated retrospective deployment trained on pre-2010 data and deployed in 2010.FindingsOur composite “rECHOmmend” model using age, sex and ECG traces had an area under the receiver operating characteristic curve (AUROC) of 0.91 and a PPV of 42% at 90% sensitivity at a prevalence of 17.9% for our composite label. Individual disease models had AUROCs ranging from 0.86-0.93 and lower PPVs from 1%-31%. The AUROC for models using different input features ranged from 0.80-0.93, increasing with additional features. Multi-site validation showed similar results to the cross-validation, with an aggregate AUROC of 0.91 across our independent test set of 11 clinical sites after training on a separate site. Our simulated retrospective deployment showed that for ECGs acquired in patients without pre-existing known structural heart disease in a single year, 2010, 11% were classified as high-risk, of which 41% developed true, echocardiography-confirmed disease within one year.InterpretationAn ECG-based machine learning model using a composite endpoint can predict previously undiagnosed, clinically significant structural heart disease while outperforming single disease models and improving practical utility with higher PPVs. This approach can facilitate targeted screening with echocardiography to improve under-diagnosis of structural heart disease.

Background: Atrial fibrillation (AF) is associated with stroke, especially when AF goes undetected. Deep neural networks (DNN) can predict incident AF from a 12-lead resting ECG. We hypothesize that use of a DNN to predict new onset AF from an ECG may identify patients at risk of sustaining a potentially preventable AF-related stroke. Methods: We trained a DNN model to predict new-onset AF using 382,604 ECGs prior to 2010. We then evaluated the model performance on a test set of ECGs from 2010 through 2014 linked to patients in an institutional stroke registry. There were 181,969 patients in the test set with at least one ECG and no prior history of AF. Of those patients 3,497 (1.9%) had a stroke following an ECG that did not show AF. Within the set of patients with stroke, 375 had the stroke within 3 years of the ECG and were diagnosed with new AF between -3 and 365 days of the stroke. We considered these potentially preventable AF-related strokes. We report the sensitivity and positive predictive value (PPV) of the model for appropriately risk stratifying these 375 patients who sustained a potentially preventable AF-related stroke. Results: We used F β scores to identify different risk prediction thresholds (operating points) for the model. Operating points chosen by F 0.5 , F 1 , and F 2 scores identified 4, 12, and 21% of the population as high risk for the development of AF within 1 year (Figure 1). Screening 1, 4, 12, and 21% of the overall population resulted in PPV of 28, 21, 15, and 12%, respectively, for identification of new onset AF in one year. Using those same thresholds yielded sensitivities of 4, 17, 45, and 62% for identifying potentially preventable AF-related strokes. The different risk prediction thresholds resulted in a low (120-162) number needed to screen to detect one potentially preventable AF-related stroke at 3 years. Conclusions: Use of a deep learning model to predict new onset AF may identify patients at high risk of sustaining a potentially preventable AF-related stroke.

Introduction: The use of convolutional neural networks (CNN) to automatically segment the heart from echocardiography images has garnered recent attention, but generalizable performance segmenting multiple cardiac structures has not been demonstrated. The objective of the present work was to develop a 2D CNN model to automatically and accurately segment the left ventricular (LV) endo- and epicardial and left atrial (LA) endocardial surfaces from an independent, external clinical dataset. Methods: A modified U-net CNN was trained using 10-fold cross-validation and augmentation within the published CAMUS echocardiography dataset to segment the LV and LA from apical 2- and 4-chamber images. For external validation, this model was applied to 3,087 echocardiograms from Geisinger for which the physician-reported LV ejection fraction (EF) was within 10% of the reported bi-plane EF, signifying confidence in quality of the underlying images. LV end-diastolic and end-systolic volumes (EDV, ESV), LA volume (LAV), and LV mass (LVM) were estimated using Simpson’s bi-plane summation. We compared performance against two published segmentation models. Results: Our model agreed well with clinically-reported values, based on small median absolute errors in percent of clinical measures (MAE; Table 1) and biases, and narrow limits of agreement (mean bias/coefficient of variation (%): LV EDV 10.3/10.6, ESV 6.3/13.1, EF 0.5/6.3, LVM 13.9/10.1, and LAV 3.1/9.5). Moreover, the observed MAE for each metric was within the previously reported limits of inter-observer variability for 2D echo. Compared to two previously published models, our model exhibited smaller MAE in all measures tested (Table 1). Conclusions: Our model exhibits accurate, generalizable performance in multi-structural echocardiography segmentation, with accuracy meeting or exceeding current leading models. Such models hold great promise for translational research and precision medicine efforts.

BackgroundAtrial fibrillation (AF) is associated with substantial morbidity, especially when it goes undetected. If new onset AF could be predicted, targeted population screening could be used to find it early. We hypothesized that a deep neural network could predict new onset AF from the resting 12-lead electrocardiogram (ECG) and that this prediction may help prevent AF-related stroke.MethodsWe used 1.6M resting 12-lead ECG voltage-time traces from 430k patients collected from 1984-2019 in this study. Deep neural networks were trained to predict new onset AF (within 1 year) in patients without a history of AF. Performance was evaluated using areas under the receiver operating characteristic curve (AUROC) and precision-recall curve (AUPRC). We performed an incidence-free survival analysis for a period of 30 years following the ECG stratified by model predictions. To simulate real-world deployment, we trained a separate model using all ECGs prior to 2010 and evaluated model performance on a test set of ECGs from 2010 through 2014 that were linked to our stroke registry. We used standard metrics to explore different prediction thresholds for the model and also calculated how many AF-related strokes might be potentially prevented.ResultsThe AUROC and AUPRC were 0.83 and 0.21, respectively, for predicting new onset AF within 1 year of an ECG. Adding age and sex improved the AUROC to 0.85 and the AUPRC to 0.23. The hazard ratio for the predicted high- vs. low-risk groups over a 30-year span was 7.2 [95% confidence interval: 6.9 – 7.6]. In a simulated deployment scenario, using the F2 score to select the risk prediction threshold, the model predicted new onset AF at 1 year with a sensitivity of 69%, specificity of 81%, and positive predictive value (PPV) of 12%. This model correctly predicted new onset AF in 62% of all patients that experienced an AF-related stroke within 3 years of the ECG.ConclusionsDeep learning can predict new onset AF from the 12-lead ECG in patients with no prior history of AF. This prediction may prove useful in preventing AF-related strokes.

Machine learning promises to assist physicians with predictions of mortality and of other future clinical events by learning complex patterns from historical data, such as longitudinal electronic health records. Here we show that a convolutional neural network trained on raw pixel data in 812,278 echocardiographic videos from 34,362 individuals provides superior predictions of one-year all-cause mortality. The model’s predictions outperformed the widely used pooled cohort equations, the Seattle Heart Failure score (measured in an independent dataset of 2,404 patients with heart failure who underwent 3,384 echocardiograms), and a machine learning model involving 58 human-derived variables from echocardiograms and 100 clinical variables derived from electronic health records. We also show that cardiologists assisted by the model substantially improved the sensitivity of their predictions of one-year all-cause mortality by 13% while maintaining prediction specificity. Large unstructured datasets may enable deep learning to improve a wide range of clinical prediction models. A deep learning model trained on raw pixel data in hundreds of thousands of echocardiographic videos for the prediction of one-year all-cause mortality outperforms clinical scores and improves predictions by cardiologists.

Segmentation of heart substructures in 2D echocardiography images is an important step in diagnosis and management of cardiovascular disease. Given the ubiquity of echocardiography in routine cardiology practice, the time-consuming nature of manual segmentation, and the high degree of inter-observer variability, fully automatic segmentation is a goal common to both clinicians and researchers. The recent publication of the annotated CA- MUS dataset will help catalyze these efforts. In this work we develop and validate against this dataset a deep fully convolutional neural network architecture for the multi-structure segmentation of echocardiography, in- cluding the left ventricular endocardium and epicardium, and the left atrium. In ten-fold cross validation with data augmentation, we obtain mean Dice overlaps of 0.93, 0.95, and 0.89 on the three structures respectively, representing state of the art on this dataset. We further report small biases and narrow limits of agreement between the automatic and manual segmentations in derived clinical indices, including median absolute errors for left ventricular diastolic (7.4mL) and systolic (4.8mL) volumes, and ejection fraction (4.1%), within previously reported inter-observer variability. These encouraging results must still be validated against large-scale independent clinical data.

Background: Arrhythmogenic right ventricular cardiomyopathy (ARVC) is associated with variants in desmosome genes. Secondary findings of pathogenic/likely pathogenic variants, primarily loss-of-function (LOF) variants, are recommended for clinical reporting; however, their prevalence and associated phenotype in a general clinical population are not fully characterized. Methods: From whole-exome sequencing of 61 019 individuals in the DiscovEHR cohort, we screened for putative loss-of-function variants in PKP2 , DSC2 , DSG2 , and DSP . We evaluated measures from prior clinical ECG and echocardiograms, manually over-read to evaluate ARVC diagnostic criteria, and performed a PheWAS (phenome-wide association study). Finally, we estimated expected penetrance using Bayesian inference. Results: One hundred forty individuals (0.23%; 59±18 years old at last encounter; 33% male) had an ARVC variant (G + ). None had an existing diagnosis of ARVC in the electronic health record, nor significant differences in prior ECG or echocardiogram findings compared with matched controls without variants. Several G + individuals satisfied major repolarization (n=4) and ventricular function (n=5) criteria, but this prevalence matched controls. PheWAS showed no significant associations of other heart disease diagnoses. Combining our best genetic and disease prevalence estimates yields an estimated penetrance of 6.0%. Conclusions: The prevalence of ARVC loss-of-function variants is ≈1:435 in a general clinical population of predominantly European descent, but with limited electronic health record-based evidence of phenotypic association in our population, consistent with a low penetrance estimate. Prospective deep phenotyping and longitudinal follow-up of a large sequenced cohort is needed to determine the true clinical relevance of an incidentally identified ARVC loss-of-function variant.

The electrocardiogram (ECG) is a widely used medical test, consisting of voltage versus time traces collected from surface recordings over the heart1. Here we hypothesized that a deep neural network (DNN) can predict an important future clinical event, 1-year all-cause mortality, from ECG voltage–time traces. By using ECGs collected over a 34-year period in a large regional health system, we trained a DNN with 1,169,662 12-lead resting ECGs obtained from 253,397 patients, in which 99,371 events occurred. The model achieved an area under the curve (AUC) of 0.88 on a held-out test set of 168,914 patients, in which 14,207 events occurred. Even within the large subset of patients (n = 45,285) with ECGs interpreted as ‘normal’ by a physician, the performance of the model in predicting 1-year mortality remained high (AUC = 0.85). A blinded survey of cardiologists demonstrated that many of the discriminating features of these normal ECGs were not apparent to expert reviewers. Finally, a Cox proportional-hazard model revealed a hazard ratio of 9.5 (P

Increased clinical use of chest high-resolution computed tomography results in increased identification of lung adenocarcinomas and persistent subsolid opacities. However, these lesions range from very indolent to extremely aggressive tumors. Clinically relevant diagnostic tools to noninvasively risk stratify and guide individualized management of these lesions are lacking. Research efforts investigating semiquantitative measures to decrease interrater and intrarater variability are emerging, and in some cases steps have been taken to automate this process. However, many such methods currently are still suboptimal, require validation and are not yet clinically applicable. The computer-aided nodule assessment and risk yield software application represents a validated tool for the automated, quantitative, and noninvasive tool for risk stratification of adenocarcinoma lung nodules. Computer-aided nodule assessment and risk yield correlates well with consensus histology and postsurgical patient outcomes, and therefore may help to guide individualized patient management, for example, in identification of nodules amenable to radiological surveillance, or in need of adjunctive therapy.

Rationale: Screening for lung cancer using low-dose computed tomography (CT) reduces lung cancer mortality. However, in addition to a high rate of benign nodules, lung cancer screening detects a large number of indolent cancers that generally belong to the adenocarcinoma spectrum. Individualized management of screen-detected adenocarcinomas would be facilitated by noninvasive risk stratification.Objectives: To validate that Computer-Aided Nodule Assessment and Risk Yield (CANARY), a novel image analysis software, successfully risk stratifies screen-detected lung adenocarcinomas based on clinical disease outcomes.Methods: We identified retrospective 294 eligible patients diagnosed with lung adenocarcinoma spectrum lesions in the low-dose CT arm of the National Lung Screening Trial. The last low-dose CT scan before the diagnosis of lung adenocarcinoma was analyzed using CANARY blinded to clinical data. Based on their parametric CANARY signatures, all the lung adenocarcinoma nodules were risk stratified into...

Pulmonary nodules are commonly detected in computed tomography (CT) chest screening of a high-risk population. The specific visual or quantitative features on CT or other modalities can be used to characterize the likelihood that a nodule is benign or malignant. Visual features on CT such as size, attenuation, location, morphology, edge characteristics, and other distinctive "signs" can be highly suggestive of a specific diagnosis and, in general, be used to determine the probability that a specific nodule is benign or malignant. Change in size, attenuation, and morphology on serial follow-up CT, or features on other modalities such as nuclear medicine studies or MRI, can also contribute to the characterization of lung nodules. Imaging analytics can objectively and reproducibly quantify nodule features on CT, nuclear medicine, and magnetic resonance imaging. Some quantitative techniques show great promise in helping to differentiate benign from malignant lesions or to stratify the risk of aggressive versus indolent neoplasm. In this article, we (1) summarize the visual characteristics, descriptors, and signs that may be helpful in management of nodules identified on screening CT, (2) discuss current quantitative and multimodality techniques that aid in the differentiation of nodules, and (3) highlight the power, pitfalls, and limitations of these various techniques.

PURPOSE Fibrotic interstitial lung diseases presenting with nonspecific and overlapping radiologic findings may be difficult to diagnose without surgical biopsy. We hypothesized that baseline quantifiable radiologic features and their short-term interval change may be predictive of underlying histologic diagnosis as well as long-term survival in idiopathic pulmonary fibrosis (IPF) presenting without honeycombing versus nonspecific interstitial pneumonia (NSIP). MATERIALS AND METHODS Forty biopsy-confirmed IPF and 20 biopsy-confirmed NSIP patients with available high-resolution chest computed tomography 4 to 24 months apart were studied. CALIPER software was used for the automated characterization and quantification of radiologic findings. RESULTS IPF subjects were older (66 vs. 48; P

High-resolution chest computed tomography (HRCT) is essential in the characterization of interstitial lung disease. The HRCT features of some diseases can be diagnostic. Longitudinal monitoring with HRCT can assess progression of interstitial lung disease; however, subtle changes in the volume and character of abnormalities can be difficult to assess. Accuracy of diagnosis can be dependent on expertise and experience of the radiologist, pathologist, or clinician. Quantitative analysis of thoracic HRCT has the potential to determine the extent of disease reproducibly, classify the types of abnormalities, and automate the diagnostic process.Novel software that utilizes histogram signatures to characterize pulmonary parenchyma was used to analyze chest HRCT data, including retrospective processing of clinical CT scans and research data from the Lung Tissue Research Consortium. Additional information including physiological, pathologic, and semiquantitative radiologist assessment was available to allow comparison of quantitative results, with visual estimates of the disease, physiological parameters, and measures of disease outcome.Quantitative analysis results were provided in regional volumetric quantities for statistical analysis and a graphical representation. These results suggest that quantitative HRCT analysis can serve as a biomarker with physiological, pathologic, and prognostic significance.It is likely that quantitative analysis of HRCT can be used in clinical practice as a means to aid in identifying a probable diagnosis, stratifying prognosis in early disease, and consistently determining progression of the disease or response to therapy. Further optimization of quantitative techniques and longitudinal analysis of well-characterized subjects would be helpful in validating these methods.

Accurate assessment of prognosis in idiopathic pulmonary fibrosis remains elusive due to significant individual radiological and physiological variability. We hypothesised that short-term radiological changes may be predictive of survival. We explored the use of CALIPER (Computer-Aided Lung Informatics for Pathology Evaluation and Rating), a novel software tool developed by the Biomedical Imaging Resource Laboratory at the Mayo Clinic Rochester (Rochester, MN, USA) for the analysis and quantification of parenchymal lung abnormalities on high-resolution computed tomography. We assessed baseline and follow-up (time-points 1 and 2, respectively) high-resolution computed tomography scans in 55 selected idiopathic pulmonary fibrosis patients and correlated CALIPER-quantified measurements with expert radiologists’ assessments and clinical outcomes. Findings of interval change (mean 289 days) in volume of reticular densities (hazard ratio 1.91, p=0.006), total volume of interstitial abnormalities (hazard ratio 1.70, p=0.003) and per cent total interstitial abnormalities (hazard ratio 1.52, p=0.017) as quantified by CALIPER were predictive of survival after a median follow-up of 2.4 years. Radiologist interpretation of short-term global interstitial lung disease progression, but not specific radiological features, was also predictive of mortality. These data demonstrate the feasibility of quantifying interval short-term changes on high-resolution computed tomography and their possible use as independent predictors of survival in idiopathic pulmonary fibrosis. Short-term quantified CT changes are predictive of survival in IPF

IntroductionPulmonary nodules of the adenocarcinoma spectrum are characterized by distinctive morphological and radiologic features and variable prognosis. Noninvasive high-resolution computed tomography–based risk stratification tools are needed to individualize their management.MethodsRadiologic measurements of histopathologic tissue invasion were developed in a training set of 54 pulmonary nodules of the adenocarcinoma spectrum and validated in 86 consecutively resected nodules. Nodules were isolated and characterized by computer-aided analysis, and data were analyzed by Spearman correlation, sensitivity, and specificity and the positive and negative predictive values.ResultsComputer-aided nodule assessment and risk yield (CANARY) can noninvasively characterize pulmonary nodules of the adenocarcinoma spectrum. Unsupervised clustering analysis of high-resolution computed tomography data identified nine unique exemplars representing the basic radiologic building blocks of these lesions. The exemplar distribution within each nodule correlated well with the proportion of histologic tissue invasion, Spearman R = 0.87, p < 0.0001 and 0.89 and p < 0.0001 for the training and the validation set, respectively. Clustering of the exemplars in three-dimensional space corresponding to tissue invasion and lepidic growth was used to develop a CANARY decision algorithm that successfully categorized these pulmonary nodules as “aggressive” (invasive adenocarcinoma) or “indolent” (adenocarcinoma in situ and minimally invasive adenocarcinoma). Sensitivity, specificity, positive predictive value, and negative predictive value of this approach for the detection of aggressive lesions were 95.4, 96.8, 95.4, and 96.8%, respectively, in the training set and 98.7, 63.6, 94.9, and 87.5%, respectively, in the validation set.ConclusionCANARY represents a promising tool to noninvasively risk stratify pulmonary nodules of the adenocarcinoma spectrum.

The aim of the study was to determine whether a novel computed tomography (CT) postprocessing software technique (CALIPER) is superior to visual CT scoring as judged by functional correlations in idiopathic pulmonary fibrosis (IPF).A total of 283 consecutive patients with IPF had CT parenchymal patterns evaluated quantitatively with CALIPER and by visual scoring. These 2 techniques were evaluated against: forced expiratory volume in 1 second (FEV1), forced vital capacity (FVC), diffusing capacity for carbon monoxide (DLco), carbon monoxide transfer coefficient (Kco), and a composite physiological index (CPI), with regard to extent of interstitial lung disease (ILD), extent of emphysema, and pulmonary vascular abnormalities.CALIPER-derived estimates of ILD extent demonstrated stronger univariate correlations than visual scores for most pulmonary function tests (PFTs): (FEV1: CALIPER R=0.29, visual R=0.18; FVC: CALIPER R=0.41, visual R=0.27; DLco: CALIPER R=0.31, visual R=0.35; CPI: CALIPER R=0.48, visual R=0.44). Correlations between CT measures of emphysema extent and PFTs were weak and did not differ significantly between CALIPER and visual scoring. Intriguingly, the pulmonary vessel volume provided similar correlations to total ILD extent scored by CALIPER for FVC, DLco, and CPI (FVC: R=0.45; DLco: R=0.34; CPI: R=0.53).CALIPER was superior to visual scoring as validated by functional correlations with PFTs. The pulmonary vessel volume, a novel CALIPER CT parameter with no visual scoring equivalent, has the potential to be a CT feature in the assessment of patients with IPF and requires further exploration.

Extracting biologically relevant data from radiology images can enable better monitoring of disease progression and therapy response. The field of radiogenomics is providing new approaches for such genomic/radiology correlations. However, there are several challenges in validation and clinical translation in that few DNA mutations are shared between tumors from different individuals and the differences in scale between imaging and genomic features can limit interpretation of underlying mechanisms. The goals of this work were to i) analyze correlations between low grade glioma (LGG) DNA somatic mutations, using a novel DNA impact scoring approach, and MRI derived imaging features; and ii) to interpret results in context of cancer hallmarks1. Multi-parametric MRI and corresponding DNA data from 32 LGG patients were extracted from The Cancer Genome Atlas (TCGA) and The Cancer Imaging Archive (TCIA). The cohort included 18 males (56%), with mean age of 44 years (range: 21-74 years). An expert radiologist outlined the normal and tumor regions of interest using ITK-Snap tool. The normal region was used as a reference to normalize image intensities in the tumor region. Tumor mean intensity and mean variance were computed from Apparent Diffusion Coefficient (ADC), T1 enhancement ratio (derived from T1 pre- and post- contract MRI), and Fluid-Attenuated Inversion Recovery (FLAIR) images. A novel algorithm was used to compute DNA impact scores for each somatic mutation. The score represents the probability of a DNA variant being pathogenic vs. nonpathogenic. First, the scoring algorithm computes a score for nucleotide base insertions, deletions, or single base changes and then computes the consequence of such changes on amino acid coding, binding sites, splice sites and protein phosphorylation sites. An impact score was then computed based on the individual DNA impact scores of mutations within the gene. Finally, an average DNA impact score was computed at the Cancer Hallmark level using a gene-cancer hallmark map. At gene level, significant positive correlations were found between the ATRX (p=0.0002), TP53 (p=0.02) and ADC mean intensity. At pathway level, regulation of TP53 expression and degradation, and DNA damage response, signal transduction by p53 class mediator, and DNA translocase activity were found to be enriched with genes that correlated with ADC and FLAIR. These pathways also contained genes that were enriched in the following cancer hallmarks: replicative immortality, evading growth suppression and genome instability. The ATRX gene is a member of all three hallmarks and TP53 a member of two. Since ADC is a measure of water diffusion and hence an indirect measure of cellularity, these findings demonstrate that mutations in replication and repair pathways are contributing to imaging features at the tumor level.1 Hanahan, D. and Weinberg, R.A. (2011). Hallmarks of cancer: the next generation. Cell 144(5):646-74. Citation Format: John Graf, Mirabela Rusu, Yunxia Sui, Dattesh Shanbhag, Uday Patil, Jeffrey Kiefer, Jill Barnholtz-Sloan, Michael Berens, Fiona Ginty, Sandeep Gupta, Chinnappa Kodira, Lee Newberg, Sushravya Raghunath, Anup Sood. Interpreting glioma MR imaging and somatic mutations in a cancer hallmark context [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 882. doi:10.1158/1538-7445.AM2017-882

Radiogenomics or radiomics is an emerging field where tumor genomic data is correlated with radiology image features, thereby potentially providing more biological information about the tumor phenotype. A central challenge is the potential for model over-fitting due to analysis of many thousands of genomic data-points with hundreds of corresponding patient image features. Biological interpretation of the imaging feature correlations is also challenged by overlapping pathways and common gene effects. Our goals were: i) to explore correlations between gene expression and corresponding Magnetic Resonance (MR) Apparent Diffusion Coefficient (ADC) derived imaging features in low grade glioma (LGG); ii) to classify significant gene and imaging correlates by cancer hallmark1. RNA expression data from 32 LGG patients were extracted from The Cancer Genome Atlas (TCGA) and matched with corresponding MR image data from The Cancer Imaging Archive (TCIA). Among 32 patients, 18 were males (56%), and ages ranged from 21 to 74 years (mean age 44). Tumor and normal regions in the MR images were annotated by an expert radiologist using ITK-Snap. The normal reference region was used normalize image intensities in corresponding tumor regions. Tumor texture features were computed on ADC Maps at each voxel location within the disease region (including first and second order statistics, Run Length and co-occurrence matrix derived measures features. The voxel features were finally aggregated within the tumor region using statistical measures of mean, variance, median, kurtosis, and skewness. ADC imaging features (n=310) were correlated with each single gene expression value (11614 genes after MAD>0.4 filtering). Only image features and genes with pairwise correlations higher than 0.68 (0.68 is the 3-standard deviation above average correlation) and FDR (False Discovery Rate) Citation Format: Yunxia Sui, Mirabela Rusu, Dattesh Shanbhag, Uday Patil, Jeffrey Kiefer, Jill Barnholtz-Sloan, Michael Berens, Fiona Ginty, Graf John, Sandeep Gupta, Chinnappa Kodira, Lee Newberg, Sushravya Raghunath, Anup Sood, Sushravya Raghunath. Elucidating cancer hallmark context from glioma MR imaging and RNA expression data [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 883. doi:10.1158/1538-7445.AM2017-883

Radiologists are adept at recognizing the character and extent of lung parenchymal abnormalities in computed tomography (CT) scans. However, the inconsistent differential diagnosis due to subjective aggregation necessitates the exploration of automated classification based on supervised or unsupervised learning. The robustness of supervised learning depends on the training samples. Towards optimizing emphysema classification, we introduce a physician-in-the-loop feedback approach to minimize ambiguity in the selected training samples. An experienced thoracic radiologist selected 412 regions of interest (ROIs) across 15 datasets to represent 124, 129, 139 and 20 training samples of mild, moderate, severe emphysema and normal appearance, respectively. Using multi-view (multiple metrics to capture complementary features) inductive learning, an ensemble of seven un-optimized support vector models (SVM) each based on a specific metric was constructed in less than 6 s. The training samples were classified using seven SVM models and consensus labels were created using majority voting. In the active relearning phase, the ensemble-expert label conflicts were resolved by the expert. The efficacy and generality of active relearning feedback was assessed in the optimized parameter space of six general purpose classifiers across the seven dissimilarity metrics. The proposed just-in-time active relearning feedback with un-optimized SVMs yielded 15 % increase in classification accuracy and 25 % reduction in the number of support vectors. The average improvement in accuracy of six classifiers in their optimized parameter space was 21 %. The proposed cooperative feedback method enhances the quality of training samples used to construct automated classification of emphysematous CT scans. Such an approach could lead to substantial improvement in quantification of emphysema.

Inter- and intra-rater agreement in assessing Interstitial Pulmonary Fibrosis (IPF) on CT scans is poor. Lack of approaches to understand the nuances of CT reconstruction parameters and their effect on patient scans hampers objective IPF quantification. In this paper, we propose an image analytic methodology to characterize the relationship between CT reconstruction parameters and IPF manifestation in scans. The raw data acquired from a patient was reconstructed with a combination of image thickness, interval and reconstruction kernels to derive 104 datasets. Quantitative and qualitative analyses were performed on the regional statistics computed in the neighborhood of each parenchymal voxel and for expert-chosen regions of interest (ROIs). The proposed approach could facilitate the selection of optimal CT reconstruction parameters to more robustly quantify IPF.

Automated lung parenchymal classification usually relies on supervised learning of expert chosen regions representative of the visually differentiable HRCT patterns specific to different pathologies (eg. emphysema, ground glass, honey combing, reticular and normal). Considering the elusiveness of a single most discriminating similarity measure, a plurality of weak learners can be combined to improve the machine learnability. Though a number of quantitative combination strategies exist, their efficacy is data and domain dependent. In this paper, we investigate multiple (N=12) quantitative consensus approaches to combine the clusters obtained with multiple (n=33) probability density-based similarity measures. Our study shows that hypergraph based meta-clustering and probabilistic clustering provides optimal expert-metric agreement.

Recently we proposed spatio-pathological stratification of lungs from multiple subjects. This enabled a pulmonary disease landscape to objectively diagnose pathology, track progression and assess pharmacologic response within and across patients. Even though the approach based on unsupervised affinity propagation clustering of a symmetric pairwise dissimilarity metric showed strong statistical and clinical correlation, it did not address the possibility of candidates being potential outliers within a cluster and consequently being triaged to suboptimal personalized care. In this paper, we address this limitation through the use of an asymmetric dissimilarity metric and a density-based outlier detection technique to identify the natural outliers within the individual clusters. In a database of 370 datasets, 28 outliers were detected among 20 clinically correlated clusters. The proposed quantitative analytics could facilitate an optimized landscape wherein every patient is triaged through the most appropriate individualized pulmonary care.

In some respects, the lung is an anatomical bog - having limited referential landmarks. Nonetheless, precise understanding of the abnormalities that inflict this organ is crucial to effective clinical diagnosis and treatment. However, wading interactively through a three-dimensional scan of the lung poses a visual quagmire to the radiologist, resulting in significant interpretive differences due to inter and intra observer variation. Despite the continuing progress in quantitative imaging, lack of unambiguous visualization with accurately, relevant cues severely hinders the clinical adoption of many computational tools. We address this unmet need through a lean visualization paradigm wherein information is presented hierarchically to provide an interactive macro-to-micro view of lung pathologies. At the macro level, the structural and functional information is summarized into a synoptic glyph that is readily interpreted and correlated to a priori known disease states. The glyphs are "patho-spatio-temporally" tagged to facilitate navigation through the level-of-detail scales, down to the micro level values in the image voxels, providing quantitative interpretation of tissue type and the confidence level in the quantitation. A novel volume compositing scheme is proposed to specify and guide to the optimal site for surgical lung biopsy. This intuitive, interactive interface for rapid and unambiguous navigation towards the clinical endpoint harnesses the power of bio-informatics technology to provide an efficient, clinically relevant and comprehensive summary of pulmonary disease, including precise location, spatial extent and intrinsic character.

Clinicians confirm the efficacy of dynamic multidisciplinary interactions in diagnosing Lung disease/wellness from CT scans. However, routine clinical practice cannot readily accomodate such interactions. Current schemes for automating lung tissue classification are based on a single elusive disease differentiating metric; this undermines their reliability in routine diagnosis. We propose a computational workflow that uses a collection (#: 15) of probability density functions (pdf)-based similarity metrics to automatically cluster pattern-specific (#patterns: 5) volumes of interest (#VOI: 976) extracted from the lung CT scans of 14 patients. The resultant clusters are refined for intra-partition compactness and subsequently aggregated into a super cluster using a cluster ensemble technique. The super clusters were validated against the consensus agreement of four clinical experts. The aggregations correlated strongly with expert consensus. By effectively mimicking the expertise of physicians, the proposed workflow could make automation of lung tissue classification a clinical reality.

Denoising is a critical preconditioning step for quantitative analysis of medical images. Despite promises for more consistent diagnosis, denoising techniques are seldom explored in clinical settings. While this may be attributed to the esoteric nature of the parameter sensitve algorithms, lack of quantitative measures on their ecacy to enhance the clinical decision making is a primary cause of physician apathy. This paper addresses this issue by exploring the eect of denoising on the integrity of supervised lung parenchymal clusters. Multiple Volumes of Interests (VOIs) were selected across multiple high resolution CT scans to represent samples of dierent patterns (normal, emphysema, ground glass, honey combing and reticular). The VOIs were labeled through consensus of four radiologists. The original datasets were ltered by multiple denoising techniques (median ltering, anisotropic diusion, bilateral ltering and non-local means) and the corresponding ltered VOIs were extracted. Plurality of cluster indices based on multiple histogram-based pair-wise similarity measures were used to assess the quality of supervised clusters in the original and ltered space. The resultant rank orders were analyzed using the Borda criteria to nd the denoising-similarity measure combination that has the best cluster quality. Our exhaustive analyis reveals (a) for a number of similarity measures, the cluster quality is inferior in the ltered space; and (b) for measures that benet from denoising, a simple median ltering outperforms non-local means and bilateral ltering. Our study suggests the need to judiciously choose, if required, a denoising technique that does not deteriorate the integrity of supervised clusters.

Radiologists are adept at recognizing the appearance of lung parenchymal abnormalities in CT scans. However, the inconsistent differential diagnosis, due to subjective aggregation, mandates supervised classification. Towards optimizing Emphysema classification, we introduce a physician-in-the-loop feedback approach in order to minimize uncertainty in the selected training samples. Using multi-view inductive learning with the training samples, an ensemble of Support Vector Machine (SVM) models, each based on a specific pair-wise dissimilarity metric, was constructed in less than six seconds. In the active relearning phase, the ensemble-expert label conflicts were resolved by an expert. This just-in-time feedback with unoptimized SVMs yielded 15% increase in classification accuracy and 25% reduction in the number of support vectors. The generality of relearning was assessed in the optimized parameter space of six different classifiers across seven dissimilarity metrics. The resultant average accuracy improved to 21%. The co-operative feedback method proposed here could enhance both diagnostic and staging throughput efficiency in chest radiology practice.

This paper introduces computational tools that could enable personalized, predictive, preemptive, and participatory (P4) Pulmonary medicine. We demonstrate approaches to (a) stratify lungs from different subjects based on the spatial distribution of parenchymal abnormality and (b) visualize the stratification through glyphs that convey both the grouping efficacy and an iconic overview of an individual's lung wellness. Affinity propagation based on regional parenchymal abnormalities is used in the referenceless stratification. Abnormalities are computed using supervised classification based on Earth Mover's distance. Twenty natural clusters were detected from 372 CT lung scans. The computed clusters correlated with clinical consensus of 9 disease types. The quality of inter- and intra-cluster stratification as assessed by ANOSIM R was 0.887 +/- 0.18 (pval0.0005). The proposed tools could serve as biomarkers to objectively diagnose pathology, track progression and assess pharmacologic response within and across patients.

This paper introduces computational tools that could enable personalized, predictive, preemptive, and participatory (P4) Pulmonary medicine. We demonstrate approaches to (a) stratify lungs from different subjects based on the spatial distribution of parenchymal abnormality and (b) visualize the stratification through glyphs that convey both the grouping efficacy and an iconic overview of an individual's lung wellness. Affinity propagation based on regional parenchymal abnormalities is used in the referenceless stratification. Abnormalities are computed using supervised classification based on Earth Mover's distance. Twenty natural clusters were detected from 372 CT lung scans. The computed clusters correlated with clinical consensus of 9 disease types. The quality of interand intra-cluster stratification as assessed by ANOSIM R was 0.887 ± 0.18 (pval < 0.0005). The proposed tools could serve as biomarkers to objectively diagnose pathology, track progression and assess pharmacologic response within and across patients.

Diffuse parenchymal lung diseases (DPLDs) are characterized by widespread pathological changes within the pulmonary tissue that impair the elasticity and gas exchange properties of the lungs. Clinical-radiological diagnosis of these diseases remains challenging and their clinical course is characterized by variable disease progression. These challenges have hindered the introduction of robust objective biomarkers for patient-specific prediction based on specific phenotypes in clinical practice for patients with DPLD. Therefore, strategies facilitating individualized clinical management, staging and identification of specific phenotypes linked to clinical disease outcomes or therapeutic responses are urgently needed. A classification schema consistently reflecting the radiological, clinical (lung function and clinical outcomes) and pathological features of a disease represents a critical need in modern pulmonary medicine. Herein, we report a quantitative stratification paradigm to identify subsets of DPLD patients with characteristic radiologic patterns in an unsupervised manner and demonstrate significant correlation of these self-organized disease groups with clinically accepted surrogate endpoints. The proposed consistent and reproducible technique could potentially transform diagnostic staging, clinical management and prognostication of DPLD patients as well as facilitate patient selection for clinical trials beyond the ability of current radiological tools. In addition, the sequential quantitative stratification of the type and extent of parenchymal process may allow standardized and objective monitoring of disease, early assessment of treatment response and mortality prediction for DPLD patients.