AI for Multistructure Incidental Findings and Mortality Prediction at Chest CT in Lung Cancer Screening.

Background Incidental extrapulmonary findings are commonly detected on chest CT scans and can be clinically important. Purpose To integrate artificial intelligence (AI)-based segmentation for multiple structures, coronary artery calcium (CAC), and epicardial adipose tissue with automated feature extraction methods and machine learning to detect extrapulmonary abnormalities and predict all-cause mortality (ACM) in a large multicenter cohort. Materials and Methods In this post hoc analysis, baseline chest CT scans in patients enrolled in the National Lung Screening Trial (NLST) from August 2002 to September 2007 were included from 33 participating sites. Per scan, 32 structures were segmented with a multistructure model. For each structure, 15 clinically interpretable radiomic features were quantified. Four general codes describing abnormalities reported by NLST radiologists were applied to identify extrapulmonary significant incidental findings on the CT scans. Death at 2-year and 10-year follow-up and the presence of extrapulmonary significant incidental findings were predicted with ensemble AI models, and individualized structure risk scores were evaluated. Area under the receiver operating characteristic curve (AUC) analysis was used to evaluate the performance of the models for prediction of ACM and extrapulmonary significant incidental findings. The Pearson χ2 test and Kruskal-Wallis rank sum test were used for statistical analyses. Results A total of 24 401 participants (median age, 61 years [IQR, 57-65 years]; 14 468 male) were included. In 3880 of 24 401 participants (16%), 4283 extrapulmonary significant incidental findings were reported. During the 10-year follow-up, 3389 of 24 401 participants (14%) died. CAC had the highest feature importance for predicting the three study end points. The 10-year ACM model demonstrated the best AUC performance (0.72; per-year mortality of 2.6% above and 0.8% below the risk threshold), followed by 2-year ACM (0.71; per-year mortality of 1.13% above and 0.3% below the risk threshold) and prediction of extrapulmonary significant incidental findings (0.70; probability of occurrence of 25.4% above and 9.6% below the threshold). Conclusion A fully automated AI model indicated extrapulmonary structures at risk on chest CT scans and predicted ACM with explanations. ClinicalTrials.gov Identifier: NCT00047385 © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Yanagawa and Hata in this issue.

CT attenuation correction improves quantitative risk prediction by cardiac SPECT in obese patients.

PURPOSE: This study aimed to compare the predictive value of CT attenuation-corrected stress total perfusion deficit (AC-sTPD) and non-corrected stress TPD (NC-sTPD) for major adverse cardiac events (MACE) in obese patients undergoing cadmium zinc telluride (CZT) SPECT myocardial perfusion imaging (MPI). METHODS: The study included 4,585 patients who underwent CZT SPECT/CT MPI for clinical indications (chest pain: 56%, shortness of breath: 13%, other: 32%) at Yale New Haven Hospital (age: 64 ± 12 years, 45% female, body mass index [BMI]: 30.0 ± 6.3 kg/m2, prior coronary artery disease: 18%). The association between AC-sTPD or NC-sTPD and MACE defined as the composite end point of mortality, nonfatal myocardial infarction or late coronary revascularization (> 90 days after SPECT) was evaluated with survival analysis. RESULTS: During a median follow-up of 25 months, 453 patients (10%) experienced MACE. In patients with BMI ≥ 35 kg/m2 (n = 931), those with AC-sTPD ≥ 3% had worse MACE-free survival than those with AC-sTPD < 3% (HR: 2.23, 95% CI: 1.40 - 3.55, p = 0.002) with no difference in MACE-free survival between patients with NC-sTPD ≥ 3% and NC-sTPD < 3% (HR:1.06, 95% CI:0.67 - 1.68, p = 0.78). AC-sTPD had higher AUC than NC-sTPD for the detection of 2-year MACE in patients with BMI ≥ 35 kg/m2 (0.631 versus 0.541, p = 0.01). In the overall cohort AC-sTPD had a higher ROC area under the curve (AUC, 0.641) than NC-sTPD (0.608; P = 0.01) for detection of 2-year MACE. In patients with BMI ≥ 35 kg/m2 AC sTPD provided significant incremental prognostic value beyond NC sTPD (net reclassification index: 0.14 [95% CI: 0.20 - 0.28]). CONCLUSIONS: AC sTPD outperformed NC sTPD in predicting MACE in patients undergoing SPECT MPI with BMI ≥ 35 kg/m2. These findings highlight the superior prognostic value of AC-sTPD in this patient population and underscore the importance of CT attenuation correction.

Unsupervised learning to characterize patients with known coronary artery disease undergoing myocardial perfusion imaging.

PURPOSE: Patients with known coronary artery disease (CAD) comprise a heterogenous population with varied clinical and imaging characteristics. Unsupervised machine learning can identify new risk phenotypes in an unbiased fashion. We use cluster analysis to risk-stratify patients with known CAD undergoing single-photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI). METHODS: From 37,298 patients in the REFINE SPECT registry, we identified 9221 patients with known coronary artery disease. Unsupervised machine learning was performed using clinical (23), acquisition (17), and image analysis (24) parameters from 4774 patients (internal cohort) and validated with 4447 patients (external cohort). Risk stratification for all-cause mortality was compared to stress total perfusion deficit (< 5%, 5-10%, ≥10%). RESULTS: Three clusters were identified, with patients in Cluster 3 having a higher body mass index, more diabetes mellitus and hypertension, and less likely to be male, have dyslipidemia, or undergo exercise stress imaging (p < 0.001 for all). In the external cohort, during median follow-up of 2.6 [0.14, 3.3] years, all-cause mortality occurred in 312 patients (7%). Cluster analysis provided better risk stratification for all-cause mortality (Cluster 3: hazard ratio (HR) 5.9, 95% confidence interval (CI) 4.0, 8.6, p < 0.001; Cluster 2: HR 3.3, 95% CI 2.5, 4.5, p < 0.001; Cluster 1, reference) compared to stress total perfusion deficit (≥10%: HR 1.9, 95% CI 1.5, 2.5 p < 0.001; < 5%: reference). CONCLUSIONS: Our unsupervised cluster analysis in patients with known CAD undergoing SPECT MPI identified three distinct phenotypic clusters and predicted all-cause mortality better than ischemia alone.

Artificial Intelligence Predicts Hospitalization for Acute Heart Failure Exacerbation in Patients Undergoing Myocardial Perfusion Imaging.

Heart failure (HF) is a leading cause of morbidity and mortality in the United States and worldwide, with a high associated economic burden. This study aimed to assess whether artificial intelligence models incorporating clinical, stress test, and imaging parameters could predict hospitalization for acute HF exacerbation in patients undergoing SPECT/CT myocardial perfusion imaging. Methods: The HF risk prediction model was developed using data from 4,766 patients who underwent SPECT/CT at a single center (internal cohort). The algorithm used clinical risk factors, stress variables, SPECT imaging parameters, and fully automated deep learning-generated calcium scores from attenuation CT scans. The model was trained and validated using repeated hold-out (10-fold cross-validation). External validation was conducted on a separate cohort of 2,912 patients. During a median follow-up of 1.9 y, 297 patients (6%) in the internal cohort were admitted for HF exacerbation. Results: The final model demonstrated a higher area under the receiver-operating-characteristic curve (0.87 ± 0.03) for predicting HF admissions than did stress left ventricular ejection fraction (0.73 ± 0.05, P < 0.0001) or a model developed using only clinical parameters (0.81 ± 0.04, P < 0.0001). These findings were confirmed in the external validation cohort (area under the receiver-operating-characteristic curve: 0.80 ± 0.04 for final model, 0.70 ± 0.06 for stress left ventricular ejection fraction, 0.72 ± 0.05 for clinical model; P < 0.001 for all). Conclusion: Integrating SPECT myocardial perfusion imaging into an artificial intelligence-based risk assessment algorithm improves the prediction of HF hospitalization. The proposed method could enable early interventions to prevent HF hospitalizations, leading to improved patient care and better outcomes.

AI-Defined Cardiac Anatomy Improves Risk Stratification of Hybrid Perfusion Imaging.

BACKGROUND: Computed tomography attenuation correction (CTAC) improves perfusion quantification of hybrid myocardial perfusion imaging by correcting for attenuation artifacts. Artificial intelligence (AI) can automatically measure coronary artery calcium (CAC) from CTAC to improve risk prediction but could potentially derive additional anatomic features. OBJECTIVES: The authors evaluated AI-based derivation of cardiac anatomy from CTAC and assessed its added prognostic utility. METHODS: The authors considered consecutive patients without known coronary artery disease who underwent single-photon emission computed tomography/computed tomography (CT) myocardial perfusion imaging at 3 separate centers. Previously validated AI models were used to segment CAC and cardiac structures (left atrium, left ventricle, right atrium, right ventricular volume, and left ventricular [LV] mass) from CTAC. They evaluated associations with major adverse cardiovascular events (MACEs), which included death, myocardial infarction, unstable angina, or revascularization. RESULTS: In total, 7,613 patients were included with a median age of 64 years. During a median follow-up of 2.4 years (IQR: 1.3-3.4 years), MACEs occurred in 1,045 (13.7%) patients. Fully automated AI processing took an average of 6.2 ± 0.2 seconds for CAC and 15.8 ± 3.2 seconds for cardiac volumes and LV mass. Patients in the highest quartile of LV mass and left atrium, LV, right atrium, and right ventricular volume were at significantly increased risk of MACEs compared to patients in the lowest quartile, with HR ranging from 1.46 to 3.31. The addition of all CT-based volumes and CT-based LV mass improved the continuous net reclassification index by 23.1%. CONCLUSIONS: AI can automatically derive LV mass and cardiac chamber volumes from CT attenuation imaging, significantly improving cardiovascular risk assessment for hybrid perfusion imaging.

Artificial Intelligence Model Predicts Sudden Cardiac Arrest Manifesting With Pulseless Electric Activity Versus Ventricular Fibrillation.

BACKGROUND: There is no specific treatment for sudden cardiac arrest (SCA) manifesting as pulseless electric activity (PEA) and survival rates are low; unlike ventricular fibrillation (VF), which is treatable by defibrillation. Development of novel treatments requires fundamental clinical studies, but access to the true initial rhythm has been a limiting factor. METHODS: Using demographics and detailed clinical variables, we trained and tested an AI model (extreme gradient boosting) to differentiate PEA-SCA versus VF-SCA in a novel setting that provided the true initial rhythm. A subgroup of SCAs are witnessed by emergency medical services personnel, and because the response time is zero, the true SCA initial rhythm is recorded. The internal cohort consisted of 421 emergency medical services-witnessed out-of-hospital SCAs with PEA or VF as the initial rhythm in the Portland, Oregon metropolitan area. External validation was performed in 220 emergency medical services-witnessed SCAs from Ventura, CA. RESULTS: In the internal cohort, the artificial intelligence model achieved an area under the receiver operating characteristic curve of 0.68 (95% CI, 0.61-0.76). Model performance was similar in the external cohort, achieving an area under the receiver operating characteristic curve of 0.72 (95% CI, 0.59-0.84). Anemia, older age, increased weight, and dyspnea as a warning symptom were the most important features of PEA-SCA; younger age, chest pain as a warning symptom and established coronary artery disease were important features associated with VF. CONCLUSIONS: The artificial intelligence model identified novel features of PEA-SCA, differentiated from VF-SCA and was successfully replicated in an external cohort. These findings enhance the mechanistic understanding of PEA-SCA with potential implications for developing novel management strategies.

Predicting mortality from AI cardiac volumes mass and coronary calcium on chest computed tomography.

Chest computed tomography is one of the most common diagnostic tests, with 15 million scans performed annually in the United States. Coronary calcium can be visualized on these scans, but other measures of cardiac risk such as atrial and ventricular volumes have classically required administration of contrast. Here we show that a fully automated pipeline, incorporating two artificial intelligence models, automatically quantifies coronary calcium, left atrial volume, left ventricular mass, and other cardiac chamber volumes in 29,687 patients from three cohorts. The model processes chamber volumes and coronary artery calcium with an end-to-end time of ~18 s, while failing to segment only 0.1% of cases. Coronary calcium, left atrial volume, and left ventricular mass index are independently associated with all-cause and cardiovascular mortality and significantly improve risk classification compared to identification of abnormalities by a radiologist. This automated approach can be integrated into clinical workflows to improve identification of abnormalities and risk stratification, allowing physicians to improve clinical decision-making.

AI-derived epicardial fat measurements improve cardiovascular risk prediction from myocardial perfusion imaging.

Epicardial adipose tissue (EAT) volume and attenuation are associated with cardiovascular risk, but manual annotation is time-consuming. We evaluated whether automated deep learning-based EAT measurements from ungated computed tomography (CT) are associated with death or myocardial infarction (MI). We included 8781 patients from 4 sites without known coronary artery disease who underwent hybrid myocardial perfusion imaging. Of those, 500 patients from one site were used for model training and validation, with the remaining patients held out for testing (n = 3511 internal testing, n = 4770 external testing). We modified an existing deep learning model to first identify the cardiac silhouette, then automatically segment EAT based on attenuation thresholds. Deep learning EAT measurements were obtained in <2 s compared to 15 min for expert annotations. There was excellent agreement between EAT attenuation (Spearman correlation 0.90 internal, 0.82 external) and volume (Spearman correlation 0.90 internal, 0.91 external) by deep learning and expert segmentation in all 3 sites (Spearman correlation 0.90-0.98). During median follow-up of 2.7 years (IQR 1.6-4.9), 565 patients experienced death or MI. Elevated EAT volume and attenuation were independently associated with an increased risk of death or MI after adjustment for relevant confounders. Deep learning can automatically measure EAT volume and attenuation from low-dose, ungated CT with excellent correlation with expert annotations, but in a fraction of the time. EAT measurements offer additional prognostic insights within the context of hybrid perfusion imaging.

Clinical phenotypes among patients with normal cardiac perfusion using unsupervised learning: a retrospective observational study.

BACKGROUND: Myocardial perfusion imaging (MPI) is one of the most common cardiac scans and is used for diagnosis of coronary artery disease and assessment of cardiovascular risk. However, the large majority of MPI patients have normal results. We evaluated whether unsupervised machine learning could identify unique phenotypes among patients with normal scans and whether those phenotypes were associated with risk of death or myocardial infarction. METHODS: Patients from a large international multicenter MPI registry (10 sites) with normal perfusion by expert visual interpretation were included in this cohort analysis. The training population included 9849 patients, and external testing population 12,528 patients. Unsupervised cluster analysis was performed, with separate training and external testing cohorts, to identify clusters, with four distinct phenotypes. We evaluated the clinical and imaging features of clusters and their associations with death or myocardial infarction. FINDINGS: Patients in Clusters 1 and 2 almost exclusively underwent exercise stress, while patients in Clusters 3 and 4 mostly required pharmacologic stress. In external testing, the risk for Cluster 4 patients (20.2% of population, unadjusted hazard ratio [HR] 6.17, 95% confidence interval [CI] 4.64-8.20) was higher than the risk associated with pharmacologic stress (HR 3.03, 95% CI 2.53-3.63), or previous myocardial infarction (HR 1.82, 95% CI 1.40-2.36). INTERPRETATION: Unsupervised learning identified four distinct phenotypes of patients with normal perfusion scans, with a significant proportion of patients at very high risk of myocardial infarction or death. Our results suggest a potential role for patient phenotyping to improve risk stratification of patients with normal imaging results. FUNDING: This work was supported by the National Heart, Lung, and Blood Institute at the National Institutes of Health [R35HL161195 to PS]. The REFINE SPECT database was supported by the National Heart, Lung, and Blood Institute at the National Institutes of Health [R01HL089765 to PS]. MCW was supported by the British Heart Foundation [FS/ICRF/20/26002].

Temporal convolutional networks and data rebalancing for clinical length of stay and mortality prediction.

It is critical for hospitals to accurately predict patient length of stay (LOS) and mortality in real-time. We evaluate temporal convolutional networks (TCNs) and data rebalancing methods to predict LOS and mortality. This is a retrospective cohort study utilizing the MIMIC-III database. The MIMIC-Extract pipeline processes 24 hour time-series clinical objective data for 23,944 unique patient records. TCN performance is compared to both baseline and state-of-the-art machine learning models including logistic regression, random forest, gated recurrent unit with decay (GRU-D). Models are evaluated for binary classification tasks (LOS > 3 days, LOS > 7 days, mortality in-hospital, and mortality in-ICU) with and without data rebalancing and analyzed for clinical runtime feasibility. Data is split temporally, and evaluations utilize tenfold cross-validation (stratified splits) followed by simulated prospective hold-out validation. In mortality tasks, TCN outperforms baselines in 6 of 8 metrics (area under receiver operating characteristic, area under precision-recall curve (AUPRC), and F-1 measure for in-hospital mortality; AUPRC, accuracy, and F-1 for in-ICU mortality). In LOS tasks, TCN performs competitively to the GRU-D (best in 6 of 8) and the random forest model (best in 2 of 8). Rebalancing improves predictive power across multiple methods and outcome ratios. The TCN offers strong performance in mortality classification and offers improved computational efficiency on GPU-enabled systems over popular RNN architectures. Dataset rebalancing can improve model predictive power in imbalanced learning. We conclude that temporal convolutional networks should be included in model searches for critical care outcome prediction systems.

On collaborative reinforcement learning to optimize the redistribution of critical medical supplies throughout the COVID-19 pandemic.

OBJECTIVE: This work investigates how reinforcement learning and deep learning models can facilitate the near-optimal redistribution of medical equipment in order to bolster public health responses to future crises similar to the COVID-19 pandemic. MATERIALS AND METHODS: The system presented is simulated with disease impact statistics from the Institute of Health Metrics, Centers for Disease Control and Prevention, and Census Bureau. We present a robust pipeline for data preprocessing, future demand inference, and a redistribution algorithm that can be adopted across broad scales and applications. RESULTS: The reinforcement learning redistribution algorithm demonstrates performance optimality ranging from 93% to 95%. Performance improves consistently with the number of random states participating in exchange, demonstrating average shortage reductions of 78.74 ± 30.8% in simulations with 5 states to 93.50 ± 0.003% with 50 states. CONCLUSIONS: These findings bolster confidence that reinforcement learning techniques can reliably guide resource allocation for future public health emergencies.