In this review, we highlight how artificial intelligence, specifically deep learning, is reshaping every aspect of cardiovascular magnetic resonance imaging: from planning and acquisition to reconstruction, analysis, and clinical report generation. We first introduce core machine learning paradigms and concepts, then survey recent deep learning advances to automate and enhance multiple aspects of MRI. We highlight the range of recent advances to provide a conceptual understanding of how the field has rapidly evolved in the last 10 years, enabling improvements in acquisition speed, spatial resolution, suppression of artifacts, and correction for motion. Automation of postprocessing is providing us a deeper look into detailed analysis of regional cardiac function and measurement of hemodynamics, and a greater ability to automatically integrate interpretation with nonimaging clinical data to support prognostication and management. Advances in artificial intelligence will continue to shape our practice of clinical cardiovascular MRI to provide greater efficiency and enrich our ability to guide the management of patients with cardiovascular disease.

In this review, we highlight how artificial intelligence, specifically deep learning, is reshaping every aspect of cardiovascular magnetic resonance imaging: from planning and acquisition to reconstruction, analysis, and clinical report generation. We first introduce core machine learning paradigms and concepts, then survey recent deep learning advances to automate and enhance multiple aspects of MRI. We highlight the range of recent advances to provide a conceptual understanding of how the field has rapidly evolved in the last 10 years, enabling improvements in acquisition speed, spatial resolution, suppression of artifacts, and correction for motion. Automation of postprocessing is providing us a deeper look into detailed analysis of regional cardiac function and measurement of hemodynamics, and a greater ability to automatically integrate interpretation with nonimaging clinical data to support prognostication and management. Advances in artificial intelligence will continue to shape our practice of clinical cardiovascular MRI to provide greater efficiency and enrich our ability to guide the management of patients with cardiovascular disease.

Purpose To develop a deep learning (DL) algorithm for identification of cardiac chamber enlargement (CCE) on anteroposterior chest radiographs using same-day transthoracic echocardiography (TTE) as a reference standard. Materials and Methods Between January 2019 and December 2021, anteroposterior chest radiographs obtained within 24 hours of TTE were retrospectively collected and randomly assigned to training (n = 5158), validation (n = 655), and test (n = 654) sets. A pretrained EfficientNet-B6 model was adapted to predict the presence of any CCE and enlargement of individual cardiac chambers. Model performance was compared with manual cardiothoracic ratio (CTR) measurements on the test set and with assessments by three cardiothoracic radiologists on a subset of 200 test set chest radiographs. Results A total of 6467 anteroposterior chest radiographs from unique patients (mean age, 63.4 years ± 17.0 [SD]; 3820 [59.1%] male patients) were included, with CCE present in 4060 (62.8%) cases. For binary classification of CCE, the model achieved an area under the receiver operating characteristic curve (AUC) value of 0.80 in the validation set and 0.83 in the test set. Corresponding performance in the validation and test sets, respectively, was as follows: accuracy, 74% and 76%; sensitivity, 76% and 81%; specificity, 71% and 66%; and area under the precision-recall curve value, 0.87 and 0.89. On the test set, the model outperformed cardiothoracic ratio (CTR) measurements (AUC, 0.83 vs 0.74) and cardiothoracic radiologist assessment (accuracy, 77% vs 60.5%-67.5%; all P < .001). Conclusion In this proof-of-concept study, a DL model demonstrated robust performance for detection of CCE on anteroposterior chest radiographs and outperformed CTR measurements and cardiothoracic radiologist assessment in a controlled evaluation setting. Keywords: Deep Learning, Algorithm Development, Neural Networks, Echocardiography, Conventional Radiography, Cardiac, Data Science, Machine Learning, Mass Chest X-Ray, Transthoracic Echocardiography, Radiographic Image Interpretation-Computer-Assisted, Cardiomegaly Supplemental material is available for this article. © RSNA, 2026.

RATIONALE Lung cancer screening (LCS) with annual low-dose computed tomography (LDCT) decreases mortality by identifying early-stage cancer but uptake is limited, especially in medically under-resourced areas. Chest x-ray tomosynthesis (CXRT) is an alternative imaging approach that can be manufactured at one sixth the cost of CT but has shown limited potential for lung cancer detection in prior studies. We investigated a novel, mobile CXRT device designed to overcome limitations of prior devices in a multireader pilot study of its performance in a composite population of patients undergoing LDCT, enriched with patients undergoing diagnostic chest CT for new lung cancer and suspicious lung nodules. METHODS: A composite population of patients undergoing screening and diagnostic chest CT between October 2023 and May 2024 were prospectively recruited to undergo imaging with next-generation CXRT device. Excluding subjects recruited for operator training, dose optimization, and radiologist training, 65 were included in a multireader study. Four subspecialty-trained cardiothoracic radiologists independently evaluated CXRT images for pulmonary lesions, blinded to patient data and CT scans. Two prespecified sets of analysis were performed to assess the accuracy of CXRT for detection of 1) biopsy-confirmed cancer and 2) pulmonary nodules on CT with ≥10 mm, ≥8 mm, and ≥6 mm long-axis size thresholds. RESULTS: Of the 65 subjects recruited for the multireader study, 20 had biopsy-proven malignancy on follow-up (median 262 days, maximum 343 days, minimum 146 days) while the remaining had benign exams. Readers showed substantial concordance (κ= 0.44-0.69) for detection of malignancy (N=20) with high sensitivity (0.85-1.0) and moderate specificity (0.56-0.60). There was moderate-high sensitivity for nodules, most of which were benign, across nodule size thresholds ≥10 mm (N=31, Sn 0.65-0.90, Sp 0.74-0.91), ≥8 mm (N=36, Sn 0.61-0.86, Sp 0.72-0.83), and ≥6 mm (N=41, Sn 0.66-0.85, Sp 0.67-0.79). CONCLUSIONS: This is the first human trial of a novel next-generation CXRT device, conducted in an at-risk pilot population enriched with participants with suspected or recurrent lung cancer. The high sensitivity of CXRT for detecting lung cancer signals its potential future use as a first-line LCS tool whereby positive screens would be referred for diagnostic CT. Notably, false negative rates with LDCT have been estimated to be 8-15% based on retrospective analysis of large LCS trials. The current multireader trial demonstrates the feasibility of CXRT for lung cancer detection and sets initial performance benchmarks to inform future studies and efforts at optimization.

Purpose:To determine whether quantitative 4-Dimensional (4D)-Flow MRI could reflect morphologic findings of pelvic venous disorder (PeVD). Methods:Abdominopelvic MRI with 4D-Flow acquired with 3T MRI from 2016-2022 were retrospectively reviewed for morphologic imaging findings: no venous abnormalities (NVA), left common iliac vein compression, left gonadal vein reflux, left renal vein (LRV) compression, and presence of pelvic collaterals. Using 4D-Flow MRI, blood flow was measured for vascular segments from the level of the suprarenal inferior vena cava (IVC) to the common iliac veins. Flow measurements at the LCIV and right common iliac vein (RCIV), the perihilar and juxta-caval renal veins were compared among participants with NVA, LCIV compression, LRV compression with and without LGV reflux, and with LGV reflux without LRV compression. Results:Sixty-six participants with LCIV compression, LRV compression, or LGV reflux displayed significantly diminished flow adjacent to the site of compression or reflux. Compared to participants with NVA, those with LCIV compression with pelvic collaterals showed increased RCIV flow and decreased LCIV flow (LCIV:RCIV flow ratio: 0.49±0.08 vs. NVA:0.92±0.05, p=0.0005). LCIV compression without pelvic collaterals did not significantly differ from NVA (LCIV:RCIV flow ratio: 0.80±0.09 vs. NVA, p>0.1). LRV compression with LGV reflux showed diminished juxta-caval LRV flow and similar perihilar LRV flow compared to LRV compression without LGV reflux ( p=0.03) or NVA (p=0.004) (juxta-caval:perihilar LRV flow ratio: with LGV reflux:0.39±0.17, without LGV reflux:1.3±0.19, NVA:1.3±0.13). Conclusions:Quantitative abdominopelvic 4D-Flow MRI measurements reflected flow diversion away from obstruction in LCIV or LRV compression, particularly in the setting of decompressing venous reservoirs.

BACKGROUND:Patients with repaired tetralogy of Fallot (rTOF) are commonly followed with cardiovascular magnetic resonance (CMR) imaging and frequently develop right ventricular (RV) dysfunction, which can be severe enough to impact left ventricular (LV) function in some patients. In this study, we sought to characterize patterns of LV dysfunction in this patient population using deep learning synthetic strain (DLSS), a fully automated deep learning algorithm capable of measuring regional LV strain and dyssynchrony. METHODS:We retrospectively collected cine steady-state free precession (SSFP) MRI images from a multi-institutional cohort of 198 patients with rTOF and 21 healthy controls. Using DLSS, we measured LV strain and strain rate across 16 American Heart Association segments from short-axis cine SSFP images and compared these values to controls. We then performed a clustering analysis to identify unique patterns of LV contraction, using segmental peak strain and several measures of dyssynchrony. We further characterized these patterns by assessing their relationship to traditional MRI metrics of volume and function. Lastly, we assessed their impact on subsequent progression to pulmonary valve replacement (PVR) through a multivariate analysis. RESULTS:Overall, patients with rTOF had decreased septal radial strain, increased lateral wall radial strain, and increased dyssynchrony relative to healthy controls. Clustering of rTOF patients identified four unique patterns of LV contraction. Most notably, patients in cluster 1 (n = 39) demonstrated an LV contraction pattern with paradoxical septal wall motion and severely reduced septal strain. These patients had significantly elevated RV end-diastolic volume relative to clusters 3 and 4 (153 ± 34 vs 127 ± 34 and 126 ± 31 mL/m2, analysis of variance p < 0.01). In the multivariate analysis, this contraction pattern was the only LV metric associated with future progression to PVR (heart rate = 2.69, p < 0.005). A smaller subset of patients (cluster 2, n = 29) showed reduced septal strain and LV ejection fraction despite synchronous ventricular contraction. CONCLUSION:Patients with rTOF demonstrate four unique patterns of LV dysfunction. Most commonly, but not exclusively, LV dysfunction is characterized by septal wall motion abnormalities and severely reduced septal strain. Patients with this pattern of LV dysfunction had concomitant RV dysfunction and rapid progression to PVR.

Purpose To define criteria for evaluating the diagnostic adequacy of expiratory CT and use these criteria to evaluate how expiratory quality affects the performance of quantitative air trapping in predicting the presence and progression of chronic lung allograft dysfunction (CLAD). Materials and Methods Consecutive post-lung transplantation inspiratory-expiratory chest CT scans acquired at the authors' institution between March 2020 and November 2023 were retrospectively evaluated for diagnostic adequacy by grading the tracheal morphology on expiratory CT scans and comparing CT lung volume measurements with spirometry data. Lung volumes and voxelwise air trapping were measured using a deep learning algorithm. Air trapping was compared against changes in spirometry data at baseline and follow-up using Pearson correlation and receiver operating characteristic curve analysis. Results A total of 603 inspiratory-expiratory chest CT scans in 192 patients who underwent lung transplantation (mean age, 57.2 years ± 13.3 [SD]); 121 male patients) were evaluated. Tracheal morphology was identified as predominantly convex on 29% (175 of 613) of the CT scans, resulting in an overestimation of the expiratory volume in these studies. The correlation of the lung volume measurements between CT and spirometry improved with tracheal concavity. A baseline air trapping level of 50% had 82.6% specificity and 34.0% sensitivity for diagnosing CLAD on studies with predominantly concave or flat morphology and a volume change of 1.0 L or greater. A 20% increase in air trapping resulted in 92.1% specificity and 20.0% sensitivity for a concurrent 10% decline in the forced expiratory volume in 1 second (FEV1). Conclusion Tracheal morphology was used to assess the diagnostic adequacy of expiratory phase CT. Increased air trapping was highly specific, but not very sensitive, for predicting an FEV1 decline and helped in the diagnosis and monitoring of CLAD progression. Keywords: CT, CT-Quantitative, Pulmonary, Lung, Physiological Studies, QA/QC, Transplantation, Technology Assessment, Quality Assurance, Artificial Intelligence, Air Trapping, Bronchiolitis Obliterans, Lung Transplant Supplemental material is available for this article. © RSNA, 2025.

Cardiac CT plays an important role in diagnosing heart diseases but is conventionally limited by its complex workflow that requires dedicated phase and bolus tracking devices [e.g., electrocardiogram (ECG) gating]. This work reports first progress towards robust and autonomous cardiac CT exams through joint deep learning (DL) and analytical analysis of pulsed-mode projections (PMPs). To this end, cardiac phase and its uncertainty were simultaneously estimated using a novel projection domain cardiac phase estimation network (PhaseNet), which utilizes sliding-window multi-channel feature extraction strategy and a long short-term memory (LSTM) block to extract temporal correlation between time-distributed PMPs. An uncertainty-driven Viterbi (UDV) regularizer was developed to refine the DL estimations at each time point through dynamic programming. Stronger regularization was performed at time points where DL estimations have higher uncertainty. The performance of the proposed phase estimation pipeline was evaluated using accurate physics-based emulated data. PhaseNet achieved improved phase estimation accuracy compared to the competing methods in terms of RMSE (~50% improvement vs. standard CNN-LSTM; ~24% improvement vs. multi-channel residual network). The added UDV regularizer resulted in an additional ~14% improvement in RMSE, achieving accurate phase estimation with <6% RMSE in cardiac phase (phase ranges from 0-100%). To our knowledge, this is the first publication of prospective cardiac phase estimation in the projection domain. Combined with our previous work on PMP-based bolus curve estimation, the proposed method could potentially be used to achieve autonomous cardiac scanning without ECG device and expert-in-the-loop bolus timing.

Foundation models (FMs) represent a transformative advancement in artificial intelligence (AI), with growing applications in medical imaging. These models leverage self-attention mechanisms and are capable of processing multimodal data, such as images, text, audio, and video, across multiple scales. Although FMs require large datasets for initial training, they can be adapted to specific medical imaging tasks using smaller labeled datasets through techniques such as transfer learning, fine-tuning, prompt engineering, few-shot learning, and zero-shot learning, making them especially valuable in data-scarce settings. Many FMs also incorporate generative AI capabilities that support the creation of synthetic medical images to further address annotation limitations. Current applications span various imaging modalities in radiology, where FMs have shown potential to improve diagnostic accuracy and streamline workflows. However, clinical integration remains challenging due to issues such as limited interpretability, potential bias, privacy concerns, regulatory constraints, high computational costs, and domain shifts between training data and real-world clinical environments. Addressing these barriers will require coordinated efforts among technical developers, health care providers, and regulatory bodies. This review explores the evolving role of FMs and generative AI in radiology, highlighting recent research advances, clinical applications, and the key challenges that must be addressed for responsible deployment.

A convolutional neural network for R-wave detection in MRI-acquired electrocardiograms achieved higher F1 scores and lower false-positive rates than standard algorithms, particularly at 3.0 T, supporting its potential to improve electrocardiographic gating and cardiac MRI scan quality.

The diagnosis of a common cause of chronic pelvic pain can be made by visualizing reflux in the ovarian veins. Fluoroscopic venography is the gold standard for diagnosing ovarian vein reflux, but it is an invasive technique that exposes patients to ionizing radiation. MRI, with its lack of ionizing radiation and capability of high-temporal and spatial-resolution vascular imaging, has the potential to provide similar diagnostic information. This retrospective report describes and assesses the utility of a dynamic contrast-enhanced MRI technique based on Differential Subsampling with Cartesian Ordering (DISCO)–MRI in 30 patients with chronic pelvic pain. Among the 14 patients who underwent both DISCO–MRI and fluoroscopic venograms, 11 (78.6%) exhibited concordant results, while 3 patients (21.4%) had discordant findings. These results suggest the potential of multiphasic contrast-enhanced DISCO–MRI as a non-invasive diagnostic tool for evaluating chronic pelvic pain.

Purpose To measure the benefit of single-phase CT, inspiratory-expiratory CT, and clinical data for convolutional neural network (CNN)-based chronic obstructive pulmonary disease (COPD) staging. Materials and Methods This retrospective study included inspiratory and expiratory lung CT images and spirometry measurements acquired between November 2007 and April 2011 from 8893 participants (mean age, 59.6 years ± 9.0 [SD]; 53.3% [4738 of 8893] male) in the COPDGene phase I cohort (ClinicalTrials.gov: NCT00608764). CNNs were trained to predict spirometry measurements (forced expiratory volume in 1 second [FEV1], FEV1 percent predicted, and ratio of FEV1 to forced vital capacity [FEV1/FVC]) using clinical data and either single-phase or multiphase CT. Spirometry predictions were then used to predict Global Initiative for Chronic Obstructive Lung Disease (GOLD) stage. Agreement between CNN-predicted and reference standard spirometry measurements and GOLD stage was assessed using intraclass correlation coefficient (ICC) and compared using bootstrapping. Accuracy for predicting GOLD stage, within-one GOLD stage, and GOLD 0 versus 1-4 was calculated. Results CNN-predicted and reference standard spirometry measurements showed moderate to good agreement (ICC, 0.66-0.79), which improved by inclusion of clinical data (ICC, 0.70-0.85; P ≤ .04), except for FEV1/FVC in the inspiratory-phase CNN model with clinical data (P = .35) and FEV1 in the expiratory-phase CNN model with clinical data (P = .33). Single-phase CNN accuracies for GOLD stage, within-one stage, and diagnosis ranged from 59.8% to 84.1% (682-959 of 1140), with moderate to good agreement (ICC, 0.68-0.70). Accuracies of CNN models using inspiratory and expiratory images ranged from 60.0% to 86.3% (684-984 of 1140), with moderate to good agreement (ICC, 0.72). Inclusion of clinical data improved agreement and accuracy for both the single-phase CNNs (ICC, 0.72; P ≤ .001; accuracy, 65.2%-85.8% [743-978 of 1140]) and inspiratory-expiratory CNNs (ICC, 0.77-0.78; P ≤ .001; accuracy, 67.6%-88.0% [771-1003 of 1140]), except expiratory CNN with clinical data (no change in GOLD stage ICC; P = .08). Conclusion CNN-based COPD diagnosis and staging using single-phase CT provides comparable accuracy with inspiratory-expiratory CT when provided clinical data relevant to staging. Keywords: Convolutional Neural Network, Chronic Obstructive Pulmonary Disease, CT, Severity Staging, Attention Map Supplemental material is available for this article. © RSNA, 2024.