Quality control-driven deep ensemble for accountable automated segmentation of cardiac magnetic resonance LGE and VNE images.

Background: Late gadolinium enhancement (LGE) cardiovascular magnetic resonance (CMR) imaging is the gold standard for non-invasive myocardial tissue characterisation. However, accurate segmentation of the left ventricular (LV) myocardium remains a challenge due to limited training data and lack of quality control. This study addresses these issues by leveraging generative adversarial networks (GAN)-generated virtual native enhancement (VNE) images to expand the training set and incorporating an automated quality control-driven (QCD) framework to improve segmentation reliability. Methods: A dataset comprising 4,716 LGE images (from 1,363 patients with hypertrophic cardiomyopathy and myocardial infarction) was used for development. To generate additional clinically validated data, LGE data were augmented with a GAN-based generator to produce VNE images. LV was contoured on these images manually by clinical observers. To create diverse candidate segmentations, the QCD framework involved multiple U-Nets, which were combined using statistical rank filters. The framework predicted the Dice Similarity Coefficient (DSC) for each candidate segmentation, with the highest predicted DSC indicating the most accurate and reliable result. The performance of the QCD ensemble framework was evaluated on both LGE and VNE test datasets (309 LGE/VNE images from 103 patients), assessing segmentation accuracy (DSC) and quality prediction (mean absolute error (MAE) and binary classification accuracy). Results: The QCD framework effectively and rapidly segmented the LV myocardium (<1 s per image) on both LGE and VNE images, demonstrating robust performance on both test datasets with similar mean DSC (LGE: 0.845±0.075; VNE: 0.845±0.071; p=ns). Incorporating GAN-generated VNE data into the training process consistently led to enhanced performance for both individual models and the overall framework. The quality control mechanism yielded a high performance (MAE=0.043, accuracy=0.951) emphasising the accuracy of the quality control-driven strategy in predicting segmentation quality in clinical settings. Overall, no statistical difference (p=ns) was found when comparing the LGE and VNE test sets across all experiments. Conclusions: The QCD ensemble framework, leveraging GAN-generated VNE data and an automated quality control mechanism, significantly improved the accuracy and reliability of LGE segmentation, paving the way for enhanced and accountable diagnostic imaging in routine clinical use.

Interpretability of radiomics models is improved when using feature group selection strategies for predicting molecular and clinical targets in clear-cell renal cell carcinoma: insights from the TRACERx Renal study.

BACKGROUND: The aim of this work is to evaluate the performance of radiomics predictions for a range of molecular, genomic and clinical targets in patients with clear cell renal cell carcinoma (ccRCC) and demonstrate the impact of novel feature selection strategies and sub-segmentations on model interpretability. METHODS: Contrast-enhanced CT scans from the first 101 patients recruited to the TRACERx Renal Cancer study (NCT03226886) were used to derive radiomics classification models to predict 20 molecular, histopathology and clinical target variables. Manual 3D segmentation was used in conjunction with automatic sub-segmentation to generate radiomics features from the core, rim, high and low enhancing sub-regions, and the whole tumour. Comparisons were made between two classification model pipelines: a Conventional pipeline reflecting common radiomics practice, and a Proposed pipeline including two novel feature selection steps designed to improve model interpretability. For both pipelines nested cross-validation was used to estimate prediction performance and tune model hyper-parameters, and permutation testing was used to evaluate the statistical significance of the estimated performance measures. Further model robustness assessments were conducted by evaluating model variability across the cross-validation folds. RESULTS: Classification performance was significant (p < 0.05, H0:AUROC = 0.5) for 11 of 20 targets using either pipeline and for these targets the AUROCs were within ± 0.05 for the two pipelines, except for one target where the Proposed pipeline performance increased by > 0.1. Five of these targets (necrosis on histology, presence of renal vein invasion, overall histological stage, linear evolutionary subtype and loss of 9p21.3 somatic alteration marker) had AUROC > 0.8. Models derived using the Proposed pipeline contained fewer feature groups than the Conventional pipeline, leading to more straightforward model interpretations without loss of performance. Sub-segmentations lead to improved performance and/or improved interpretability when predicting the presence of sarcomatoid differentiation and tumour stage. CONCLUSIONS: Use of the Proposed pipeline, which includes the novel feature selection methods, leads to more interpretable models without compromising prediction performance. TRIAL REGISTRATION: NCT03226886 (TRACERx Renal).

Incident Clinical and Mortality Associations of Myocardial Native T1 in the UK Biobank.

BACKGROUND: Cardiac magnetic resonance native T1-mapping provides noninvasive, quantitative, and contrast-free myocardial characterization. However, its predictive value in population cohorts has not been studied. OBJECTIVES: The associations of native T1 with incident events were evaluated in 42,308 UK Biobank participants over 3.17 ± 1.53 years of prospective follow-up. METHODS: Native T1-mapping was performed in 1 midventricular short-axis slice using the Shortened Modified Look-Locker Inversion recovery technique (WIP780B) in 1.5-T scanners (Siemens Healthcare). Global myocardial T1 was calculated using an automated tool. Associations of T1 with: 1) prevalent risk factors (eg, diabetes, hypertension, and high cholesterol); 2) prevalent and incident diseases (eg, any cardiovascular disease [CVD], any brain disease, valvular heart disease, heart failure, nonischemic cardiomyopathies, cardiac arrhythmias, atrial fibrillation [AF], myocardial infarction, ischemic heart disease [IHD], and stroke); and 3) mortality (eg, all-cause, CVD, and IHD) were examined. Results are reported as odds ratios (ORs) or HRs per SD increment of T1 value with 95% CIs and corrected P values, from logistic and Cox proportional hazards regression models. RESULTS: Higher myocardial T1 was associated with greater odds of a range of prevalent conditions (eg, any CVD, brain disease, heart failure, nonischemic cardiomyopathies, AF, stroke, and diabetes). The strongest relationships were with heart failure (OR: 1.41 [95% CI: 1.26-1.57]; P = 1.60 × 10-9) and nonischemic cardiomyopathies (OR: 1.40 [95% CI: 1.16-1.66]; P = 2.42 × 10-4). Native T1 was positively associated with incident AF (HR: 1.25 [95% CI: 1.10-1.43]; P = 9.19 × 10-4), incident heart failure (HR: 1.47 [95% CI: 1.31-1.65]; P = 4.79 × 10-11), all-cause mortality (HR: 1.24 [95% CI: 1.12-1.36]; P = 1.51 × 10-5), CVD mortality (HR: 1.40 [95% CI: 1.14-1.73]; P = 0.0014), and IHD mortality (HR: 1.36 [95% CI: 1.03-1.80]; P = 0.0310). CONCLUSIONS: This large population study demonstrates the utility of myocardial native T1-mapping for disease discrimination and outcome prediction.

Acute Response in the Noninfarcted Myocardium Predicts Long-Term Major Adverse Cardiac Events After STEMI.

BACKGROUND: Acute ST-segment elevation myocardial infarction (STEMI) has effects on the myocardium beyond the immediate infarcted territory. However, pathophysiologic changes in the noninfarcted myocardium and their prognostic implications remain unclear. OBJECTIVES: The purpose of this study was to evaluate the long-term prognostic value of acute changes in both infarcted and noninfarcted myocardium post-STEMI. METHODS: Patients with acute STEMI undergoing primary percutaneous coronary intervention underwent evaluation with blood biomarkers and cardiac magnetic resonance (CMR) at 2 days and 6 months, with long-term follow-up for major adverse cardiac events (MACE). A comprehensive CMR protocol included cine, T2-weighted, T2∗, T1-mapping, and late gadolinium enhancement (LGE) imaging. Areas without LGE were defined as noninfarcted myocardium. MACE was a composite of cardiac death, sustained ventricular arrhythmia, and new-onset heart failure. RESULTS: Twenty-two of 219 patients (10%) experienced an MACE at a median of 4 years (IQR: 2.5-6.0 years); 152 patients returned for the 6-month visit. High T1 (>1250 ms) in the noninfarcted myocardium was associated with lower left ventricular ejection fraction (LVEF) (51% ± 8% vs 55% ± 9%; P = 0.002) and higher NT-pro-BNP levels (290 pg/L [IQR: 103-523 pg/L] vs 170 pg/L [IQR: 61-312 pg/L]; P = 0.008) at 6 months and a 2.5-fold (IQR: 1.03-6.20) increased risk of MACE (2.53 [IQR: 1.03-6.22]), compared with patients with normal T1 in the noninfarcted myocardium (P = 0.042). A lower T1 (<1,300 ms) in the infarcted myocardium was associated with increased MACE (3.11 [IQR: 1.19-8.13]; P = 0.020). Both noninfarct and infarct T1 were independent predictors of MACE (both P = 0.001) and significantly improved risk prediction beyond LVEF, infarct size, and microvascular obstruction (C-statistic: 0.67 ± 0.07 vs 0.76 ± 0.06, net-reclassification index: 40% [IQR: 12%-64%]; P = 0.007). CONCLUSIONS: The acute responses post-STEMI in both infarcted and noninfarcted myocardium are independent incremental predictors of long-term MACE. These insights may provide new opportunities for treatment and risk stratification in STEMI.

Cardiac stress T1-mapping response and extracellular volume stability of MOLLI-based T1-mapping methods.

Stress and rest T1-mapping may assess for myocardial ischemia and extracellular volume (ECV). However, the stress T1 response is method-dependent, and underestimation may lead to misdiagnosis. Further, ECV quantification may be affected by time, as well as the number and dosage of gadolinium (Gd) contrast administered. We compared two commonly available T1-mapping approaches in their stress T1 response and ECV measurement stability. Healthy subjects (n = 10, 50% female, 35 ± 8 years) underwent regadenoson stress CMR (1.5 T) on two separate days. Prototype ShMOLLI 5(1)1(1)1 sequence was used to acquire consecutive mid-ventricular T1-maps at rest, stress and post-Gd contrast to track the T1 time evolution. For comparison, standard MOLLI sequences were used: MOLLI 5(3)3 Low (256 matrix) & High (192 matrix) Heart Rate (HR) to acquire rest and stress T1-maps, and MOLLI 4(1)3(1)2 Low & High HR for post-contrast T1-maps. Stress and rest myocardial blood flow (MBF) maps were acquired after IV Gd contrast (0.05 mmol/kg each). Stress T1 reactivity (delta T1) was defined as the relative percentage increase in native T1 between rest and stress. Myocardial T1 values for delta T1 (dT1) and ECV were calculated. Residuals from the identified time dependencies were used to assess intra-method variability. ShMOLLI achieved a greater stress T1 response compared to MOLLI Low and High HR (peak dT1 = 6.4 ± 1.7% vs. 4.8 ± 1.3% vs. 3.8 ± 1.0%, respectively; both p < 0.0001). ShMOLLI dT1 correlated strongly with stress MBF (r = 0.77, p < 0.001), compared to MOLLI Low HR (r = 0.65, p < 0.01) and MOLLI High HR (r = 0.43, p = 0.07). ShMOLLI ECV was more stable to gadolinium dose with less time drift (0.006-0.04% per minute) than MOLLI variants. Overall, ShMOLLI demonstrated less intra-individual variability than MOLLI variants for stress T1 and ECV quantification. Power calculations indicate up to a fourfold (stress T1) and 7.5-fold (ECV) advantage in sample-size reduction using ShMOLLI. Our results indicate that ShMOLLI correlates strongly with increased MBF during regadenoson stress and achieves a significantly higher stress T1 response, greater effect size, and greater ECV measurement stability compared with the MOLLI variants tested.

Toward Replacing Late Gadolinium Enhancement With Artificial Intelligence Virtual Native Enhancement for Gadolinium-Free Cardiovascular Magnetic Resonance Tissue Characterization in Hypertrophic Cardiomyopathy.

BACKGROUND: Late gadolinium enhancement (LGE) cardiovascular magnetic resonance (CMR) imaging is the gold standard for noninvasive myocardial tissue characterization but requires intravenous contrast agent administration. It is highly desired to develop a contrast agent-free technology to replace LGE for faster and cheaper CMR scans. METHODS: A CMR virtual native enhancement (VNE) imaging technology was developed using artificial intelligence. The deep learning model for generating VNE uses multiple streams of convolutional neural networks to exploit and enhance the existing signals in native T1 maps (pixel-wise maps of tissue T1 relaxation times) and cine imaging of cardiac structure and function, presenting them as LGE-equivalent images. The VNE generator was trained using generative adversarial networks. This technology was first developed on CMR datasets from the multicenter Hypertrophic Cardiomyopathy Registry, using hypertrophic cardiomyopathy as an exemplar. The datasets were randomized into 2 independent groups for deep learning training and testing. The test data of VNE and LGE were scored and contoured by experienced human operators to assess image quality, visuospatial agreement, and myocardial lesion burden quantification. Image quality was compared using a nonparametric Wilcoxon test. Intra- and interobserver agreement was analyzed using intraclass correlation coefficients (ICC). Lesion quantification by VNE and LGE were compared using linear regression and ICC. RESULTS: A total of 1348 hypertrophic cardiomyopathy patients provided 4093 triplets of matched T1 maps, cines, and LGE datasets. After randomization and data quality control, 2695 datasets were used for VNE method development and 345 were used for independent testing. VNE had significantly better image quality than LGE, as assessed by 4 operators (n=345 datasets; P<0.001 [Wilcoxon test]). VNE revealed lesions characteristic of hypertrophic cardiomyopathy in high visuospatial agreement with LGE. In 121 patients (n=326 datasets), VNE correlated with LGE in detecting and quantifying both hyperintensity myocardial lesions (r=0.77-0.79; ICC=0.77-0.87; P<0.001) and intermediate-intensity lesions (r=0.70-0.76; ICC=0.82-0.85; P<0.001). The native CMR images (cine plus T1 map) required for VNE can be acquired within 15 minutes and producing a VNE image takes less than 1 second. CONCLUSIONS: VNE is a new CMR technology that resembles conventional LGE but without the need for contrast administration. VNE achieved high agreement with LGE in the distribution and quantification of lesions, with significantly better image quality.

Deep neural network ensemble for on-the-fly quality control-driven segmentation of cardiac MRI T1 mapping.

Recent developments in artificial intelligence have generated increasing interest to deploy automated image analysis for diagnostic imaging and large-scale clinical applications. However, inaccuracy from automated methods could lead to incorrect conclusions, diagnoses or even harm to patients. Manual inspection for potential inaccuracies is labor-intensive and time-consuming, hampering progress towards fast and accurate clinical reporting in high volumes. To promote reliable fully-automated image analysis, we propose a quality control-driven (QCD) segmentation framework. It is an ensemble of neural networks that integrate image analysis and quality control. The novelty of this framework is the selection of the most optimal segmentation based on predicted segmentation accuracy, on-the-fly. Additionally, this framework visualizes segmentation agreement to provide traceability of the quality control process. In this work, we demonstrated the utility of the framework in cardiovascular magnetic resonance T1-mapping - a quantitative technique for myocardial tissue characterization. The framework achieved near-perfect agreement with expert image analysts in estimating myocardial T1 value (r=0.987,p<.0005; mean absolute error (MAE)=11.3ms), with accurate segmentation quality prediction (Dice coefficient prediction MAE=0.0339) and classification (accuracy=0.99), and a fast average processing time of 0.39 second/image. In summary, the QCD framework can generate high-throughput automated image analysis with speed and accuracy that is highly desirable for large-scale clinical applications.

Cardiovascular magnetic resonance stress and rest T1-mapping using regadenoson for detection of ischemic heart disease compared to healthy controls.

BACKGROUND: Adenosine stress T1-mapping on cardiovascular magnetic resonance (CMR) can differentiate between normal, ischemic, infarcted, and remote myocardial tissue classes without the need for contrast agents. Regadenoson, a selective coronary vasodilator, is often used in stress perfusion imaging when adenosine is contra-indicated, and has advantages in ease of administration, safety profile, and clinical workflow. We aimed to characterize the regadenoson stress T1-mapping response in healthy individuals, and to investigate its ability to differentiate between myocardial tissue classes in patients with coronary artery disease (CAD). METHODS: Eleven healthy controls and 25 patients with CAD underwent regadenoson stress perfusion CMR, as well as rest and stress ShMOLLI T1-mapping. Native T1 values and stress T1 reactivity were derived for normal myocardium in healthy controls and for different myocardial tissue classes in patients with CAD. RESULTS: Healthy controls had normal myocardial native T1 values at rest (931 ± 22 ms) with significant global regadenoson stress T1 reactivity (δT1 = 8.2 ± 0.8% relative to baseline; p < 0.0001). Infarcted myocardium had significantly higher resting T1 (1215 ± 115 ms) than ischemic, remote, and normal myocardium (all p < 0.0001) with an abolished stress T1 response (δT1 = -0.8% [IQR: -1.9-0.5]). Ischemic myocardium had elevated resting T1 compared to normal (964 ± 57 ms; p < 0.01) with an abolished stress T1 response (δT1 = 0.5 ± 1.6%). Remote myocardium in patients had comparable resting T1 to normal (949 ms [IQR: 915-973]; p = 0.06) with blunted stress reactivity (δT1 = 4.3% [IQR: 3.1-6.3]; p < 0.0001). CONCLUSIONS: Healthy controls demonstrate significant stress T1 reactivity during regadenoson stress. Regadenoson stress and rest T1-mapping is a viable alternative to adenosine and exercise for the assessment of CAD and can distinguish between normal, ischemic, infarcted, and remote myocardium.

Standardization of T1-mapping in cardiovascular magnetic resonance using clustered structuring for benchmarking normal ranges.

BACKGROUND: Cardiovascular magnetic resonance T1-mapping is increasingly used for tissue characterization, commonly based on Modified Look-Locker Inversion recovery (MOLLI). However, there are numerous MOLLI variants with differing normal ranges. This lack of standardization presents confusion and difficulty in inter-center comparisons, hindering widespread adoption of T1-mapping. METHODS: To address this, we performed a structured literature search for native left ventricular myocardial T1-mapping in healthy humans measured using MOLLI variants at 1.5 and 3 Tesla, across scanner vendors. We then used k-means clustering to structure normal MOLLI-T1 values according to magnetic field strength, and investigated correlations between common imaging parameters: repetition time (TR), echo time (TE), flip angle (FA). RESULTS: We analyzed data from 2207 healthy controls in 76 independent reports. Normal MOLLI-T1 standard deviations varied by 11-fold, and dependencies on TE, TR, and FA differed between 1.5 T and 3 T, thwarting meaningful T1 standardization even within a single field strength, including the use of Z-score. However, divergent MOLLI-T1 norms may be structured using data clustering. For 1.5 T, two clusters emerged: Cluster11.5T: T1 = 958 ± 16 ms (n = 1280); Cluster21.5T: T1 = 1027 ± 19 ms (n = 386). For 3 T, three clusters emerged: Cluster13T: T1 = 1160 ± 21 ms (n = 330); Cluster23T: T1 = 1067 ± 18 ms (n = 178); Cluster33T: T1 = 1227 ± 19 ms (n = 41). We then propose the concept of an online calculator for assigning local norms to a known MOLLI-T1 cluster, allowing benchmarking against published norms. CONCLUSION: Clustered structuring allows T1 standardization of widely-divergent MOLLI variants, benchmarking local norms (usually based on smaller samples) against published norms (larger samples). This may increase confidence and quality control in method implementation, facilitating wider clinical adoption of T1-mapping.

Deep learning with attention supervision for automated motion artefact detection in quality control of cardiac T1-mapping.

Cardiac magnetic resonance quantitative T1-mapping is increasingly used for advanced myocardial tissue characterisation. However, cardiac or respiratory motion can significantly affect the diagnostic utility of T1-maps, and thus motion artefact detection is critical for quality control and clinically-robust T1 measurements. Manual quality control of T1-maps may provide reassurance, but is laborious and prone to error. We present a deep learning approach with attention supervision for automated motion artefact detection in quality control of cardiac T1-mapping. Firstly, we customised a multi-stream Convolutional Neural Network (CNN) image classifier to streamline the process of automatic motion artefact detection. Secondly, we imposed attention supervision to guide the CNN to focus on targeted myocardial segments. Thirdly, when there was disagreement between the human operator and machine, a second human validator reviewed and rescored the cases for adjudication and to identify the source of disagreement. The multi-stream neural networks demonstrated 89.8% agreement, 87.4% ROC-AUC on motion artefact detection with the human operator in the 2568 T1 maps. Trained with additional supervision on attention, agreements and AUC significantly improved to 91.5% and 89.1%, respectively (p < 0.001). Rescoring of disagreed cases by the second human validator revealed that human operator error was the primary cause of disagreement. Deep learning with attention supervision provides a quick and high-quality assurance of clinical images, and outperforms human operators.

Automated localization and quality control of the aorta in cine CMR can significantly accelerate processing of the UK Biobank population data.

INTRODUCTION: Aortic distensibility can be calculated using semi-automated methods to segment the aortic lumen on cine CMR (Cardiovascular Magnetic Resonance) images. However, these methods require visual quality control and manual localization of the region of interest (ROI) of ascending (AA) and proximal descending (PDA) aorta, which limit the analysis in large-scale population-based studies. Using 5100 scans from UK Biobank, this study sought to develop and validate a fully automated method to 1) detect and locate the ROIs of AA and PDA, and 2) provide a quality control mechanism. METHODS: The automated AA and PDA detection-localization algorithm followed these steps: 1) foreground segmentation; 2) detection of candidate ROIs by Circular Hough Transform (CHT); 3) spatial, histogram and shape feature extraction for candidate ROIs; 4) AA and PDA detection using Random Forest (RF); 5) quality control based on RF detection probability. To provide the ground truth, overall image quality (IQ = 0-3 from poor to good) and aortic locations were visually assessed by 13 observers. The automated algorithm was trained on 1200 scans and Dice Similarity Coefficient (DSC) was used to calculate the agreement between ground truth and automatically detected ROIs. RESULTS: The automated algorithm was tested on 3900 scans. Detection accuracy was 99.4% for AA and 99.8% for PDA. Aorta localization showed excellent agreement with the ground truth, with DSC ≥ 0.9 in 94.8% of AA (DSC = 0.97 ± 0.04) and 99.5% of PDA cases (DSC = 0.98 ± 0.03). AA×PDA detection probabilities could discriminate scans with IQ ≥ 1 from those severely corrupted by artefacts (AUC = 90.6%). If scans with detection probability < 0.75 were excluded (350 scans), the algorithm was able to correctly detect and localize AA and PDA in all the remaining 3550 scans (100% accuracy). CONCLUSION: The proposed method for automated AA and PDA localization was extremely accurate and the automatically derived detection probabilities provided a robust mechanism to detect low quality scans for further human review. Applying the proposed localization and quality control techniques promises at least a ten-fold reduction in human involvement without sacrificing any accuracy.