An artificial intelligence tool for automated analysis of large-scale unstructured clinical cine cardiac magnetic resonance databases

Abstract

Aims: Artificial intelligence (AI) techniques have been proposed for automating analysis of short-axis (SAX) cine cardiac magnetic resonance (CMR), but no CMR analysis tool exists to automatically analyse large (unstructured) clinical CMR datasets. We develop and validate a robust AI tool for start-to-end automatic quantification of cardiac function from SAX cine CMR in large clinical databases.

Methods and results: Our pipeline for processing and analysing CMR databases includes automated steps to identify the correct data, robust image pre-processing, an AI algorithm for biventricular segmentation of SAX CMR and estimation of functional biomarkers, and automated post-analysis quality control to detect and correct errors. The segmentation algorithm was trained on 2793 CMR scans from two NHS hospitals and validated on additional cases from this dataset (n = 414) and five external datasets (n = 6888), including scans of patients with a range of diseases acquired at 12 different centres using CMR scanners from all major vendors. Median absolute errors in cardiac biomarkers were within the range of inter-observer variability: <8.4 mL (left ventricle volume), <9.2 mL (right ventricle volume), <13.3 g (left ventricular mass), and <5.9% (ejection fraction) across all datasets. Stratification of cases according to phenotypes of cardiac disease and scanner vendors showed good performance across all groups.

Conclusion: We show that our proposed tool, which combines image pre-processing steps, a domain-generalizable AI algorithm trained on a large-scale multi-domain CMR dataset and quality control steps, allows robust analysis of (clinical or research) databases from multiple centres, vendors, and cardiac diseases. This enables translation of our tool for use in fully automated processing of large multi-centre databases.

Keywords: Artificial intelligence; Cardiac function; Cardiac magnetic resonance; Cardiac segmentation; Quality control.

Graphical Abstract

AI-based, quality-controlled CMR analysis tool. The proposed CMR analysis tool consists of CMR image pre-processing steps, an AI method that automatically selects the cine acquisitions prior to image analysis, an AI method that segments the ventricles and the myocardium from short-axis cine CMR stacks, and a post-analysis QC step. QAgt, ground truth segmentation data quality assessment; QC, quality control.

Figure 1

Bland–Altman analysis of cardiac volume, ejection fraction, and mass: Cardiac biomarkers derived from manual and automated segmentations were compared for all validation cases. The thick line depicts the mean bias between the automated and manual analyses. The top and bottom dotted lines correspond to +1.96 and −1.96 standard deviations from the mean bias, respectively. The Pearson’s correlation coefficients (R) between our method and the manual analysis (and the corresponding P-values) are indicated for each cardiac biomarker. LVEDV, left ventricular end-diastolic volume; LVESV, left ventricular end-systolic volume; LVEF, left ventricular ejection fraction; LVM, left ventricular mass; RVEDV, right ventricular end-diastolic volume; RVEF, right ventricular ejection fraction; RVESV, right ventricular end-systolic volume.

Figure 2

Box plots of manually and automatically derived cardiac biomarkers for each disease group: Statistical differences from zero were assessed using Wilcoxon signed-rank tests. Pairwise post hoc testing was performed using Bonferroni correction for multiple comparisons. Asterisks indicate statistically significant differences from zero for each group after correction (five tests), where *P < 0.01/5, **P < 0.001/5, and ***P < 0.0001/5. CHD, congenital heart disease; DCM, dilated cardiomyopathy; IHD, ischaemic heart disease; LVEDV, left ventricular end-diastolic volume; LVESV, left ventricular end-systolic volume; LVEF, left ventricular ejection fraction; LVM, left ventricular mass; NOR, normal cardiac anatomy and function; Other, other cardiac diseases; RVEDV, right ventricular end-diastolic volume; RVEF, right ventricular ejection fraction; RVESV, right ventricular end-systolic volume.

Figure 3

Box plots of manually and automatically derived cardiac biomarkers for each scanner group: Statistical differences from zero were assessed using Wilcoxon signed-rank tests. Pairwise post hoc testing was performed using Bonferroni correction for multiple comparisons. Asterisks indicate statistically significant differences from zero for each group after correction (four tests), where *P < 0.01/4, **P < 0.001/4, and ***P < 0.0001/4. CHD, congenital heart disease; DCM, dilated cardiomyopathy; IHD, ischaemic heart disease; LVEDV, left ventricular end-diastolic volume; LVESV, left ventricular end-systolic volume; LVEF, left ventricular ejection fraction; LVM, left ventricular mass; NOR, normal cardiac anatomy and function; Other, other cardiac diseases; RVEDV, right ventricular end-diastolic volume; RVEF, right ventricular ejection fraction; RVESV, right ventricular end-systolic volume.

Figure 4

Examples of automated segmentations from different disease groups: The middle slice of each automated segmentation is shown in end-diastole (ED) and end-systole (ES). Cardiac labels: LV blood pool (red), LV myocardium (yellow), and RV blood pool (blue).

Figure 5

Examples of erroneous ground truth segmentations identified during manual QA^gt. Cardiac labels: LV blood pool (red), LV myocardium (yellow) and RV blood pool (blue). (A) Image in ED: note that the LV myocardium is segmented but the LV blood pool segmentation is absent and that the RV segmentation is labelled as myocardium (yellow); (B) top slice of the cine stack in ED: the basal part of the heart is not included in the cine SAX stack; (C) image in ED: note the unusual LV structure that was segmented and the absence of an RV segmentation; (D) image in ES; note the absence of LV and RV segmentations, while myocardium is present for both.