A deep learning pipeline for automatic analysis of multi-scan cardiovascular magnetic resonance.

BACKGROUND: Cardiovascular magnetic resonance (CMR) sequences are commonly used to obtain a complete description of the function and structure of the heart, provided that accurate measurements are extracted from images. New methods of extraction of information are being developed, among them, deep neural networks are powerful tools that showed the ability to perform fast and accurate segmentation. Iq1n order to reduce the time spent by reading physicians to process data and minimize intra- and inter-observer variability, we propose a fully automatic multi-scan CMR image analysis pipeline. METHODS: Sequence specific U-Net 2D models were trained to perform the segmentation of the left ventricle (LV), right ventricle (RV) and aorta in cine short-axis, late gadolinium enhancement (LGE), native T1 map, post-contrast T1, native T2 map and aortic flow sequences depending on the need. The models were trained and tested on a set of data manually segmented by experts using semi-automatic and manual tools. A set of parameters were computed from the resulting segmentations such as the left ventricular and right ventricular ejection fraction (EF), LGE scar percentage, the mean T1, T1 post, T2 values within the myocardium, and aortic flow. The Dice similarity coefficient, Hausdorff distance, mean surface distance, and Pearson correlation coefficient R were used to assess and compare the results of the U-Net based pipeline with intra-observer variability. Additionally, the pipeline was validated on two clinical studies. RESULTS: The sequence specific U-Net 2D models trained achieved fast (≤ 0.2 s/image on GPU) and precise segmentation over all the targeted region of interest with high Dice scores (= 0.91 for LV, = 0.92 for RV, = 0.93 for Aorta in average) comparable to intra-observer Dice scores (= 0.86 for LV, = 0.87 for RV, = 0.95 for aorta flow in average). The automatically and manually computed parameters were highly correlated (R = 0.91 in average) showing results superior to the intra-observer variability (R = 0.85 in average) for every sequence presented here. CONCLUSION: The proposed pipeline allows for fast and robust analysis of large CMR studies while guaranteeing reproducibility, hence potentially improving patient's diagnosis as well as clinical studies outcome.

Cardiac motion and spillover correction for quantitative PET imaging using dynamic MRI.

PURPOSE: Cardiac positron emission tomography/magnetic resonance imaging (PET/MRI) acquisition presents novel clinical applications thanks to the combination of viability and metabolic imaging (PET) and functional and structural imaging (MRI). However, the resolution of PET, as well as cardiac and respiratory motion in nongated cardiac imaging acquisition protocols, leads to a reduction in image quality and severe quantitative bias. Respiratory or cardiac motion is customarily addressed with gated reconstruction which results in higher noise. METHODS: Inspired by a method that has been used in brain PET, a practical correction approach, designed to overcome these existing limitations for quantitative PET imaging, was developed and applied in the context of cardiac PET/MRI. The correction approach for PET data consists of computing the mean density map of each underlying moving region, as obtained with MRI, and translating them to the PET space taking into account the PET spatial and temporal resolution. Using these tissue density maps, the method then constructs a system of linear equations that models the activity recovery and cross-contamination coefficients, which can be solved for the true activity values. Physical and numerical cardiac phantoms were employed in order to quantify the proposed correction. The full correction pipeline was then used to assess differences in metabolic function between scar and healthy myocardium in eight patients with recent acute myocardial infarction using [11 C]-acetate. Data from ten additional patients, injected with [18 F]-FDG, were used to compare the method to the standard electrocardiography (ECG)-gated approach. RESULTS: The proposed method resulted in better recovery (from 32% to 95% on the simulated phantom model) and less residual activity than the standard approach. Higher signal-to-noise and contrast-to-noise ratios than ECG-gating were also witnessed (Signal-to-noise ratio (SNR) increased from 2.92 to 5.24, contrast-to-noise ratio (CNR) increased from 62.9 to 145.9 when compared to a four-gate reconstruction). Finally, the relevance of this correction using [11 C]-acetate PET patient data, for which erroneous physiological conclusions could have been made based on the uncorrected data, was established as the correction led to the expected clinical results. CONCLUSIONS: An efficient and simple method to correct for the quantitative biases in PET measurements caused by cardiac motion has been developed. Validation experiments using phantom and patient data showed improved accuracy and reliability with this approach when compared to simpler strategies such as gated acquisition or optimal regions of interest (ROI).