Clinical assessment of an AI tool for measuring biventricular parameters on cardiac MR

Introduction

Cardiac magnetic resonance (CMR) is of diagnostic and prognostic value in a range of cardiopulmonary conditions. Current methods for evaluating CMR studies are laborious and time-consuming, contributing to delays for patients. As the demand for CMR increases, there is a growing need to automate this process. The application of artificial intelligence (AI) to CMR is promising, but the evaluation of these tools in clinical practice has been limited. This study assessed the clinical viability of an automatic tool for measuring cardiac volumes on CMR.

Methods

Consecutive patients who underwent CMR for any indication between January 2022 and October 2022 at a single tertiary centre were included prospectively. For each case, short-axis CMR images were segmented by the AI tool and manually to yield volume, mass and ejection fraction measurements for both ventricles. Automated and manual measurements were compared for agreement and the quality of the automated contours was assessed visually by cardiac radiologists.

Results

462 CMR studies were included. No statistically significant difference was demonstrated between any automated and manual measurements (p > 0.05; independent T-test). Intraclass correlation coefficient and Bland-Altman analysis showed excellent agreement across all metrics (ICC > 0.85). The automated contours were evaluated visually in 251 cases, with agreement or minor disagreement in 229 cases (91.2%) and failed segmentation in only a single case (0.4%). The AI tool was able to provide automated contours in under 90 s.

Conclusions

Automated segmentation of both ventricles on CMR by an automatic tool shows excellent agreement with manual segmentation performed by CMR experts in a retrospective real-world clinical cohort. Implementation of the tool could improve the efficiency of CMR reporting and reduce delays between imaging and diagnosis.

Training and clinical testing of artificial intelligence derived right atrial cardiovascular magnetic resonance measurements

BACKGROUND

Right atrial (RA) area predicts mortality in patients with pulmonary hypertension, and is recommended by the European Society of Cardiology/European Respiratory Society pulmonary hypertension guidelines. The advent of deep learning may allow more reliable measurement of RA areas to improve clinical assessments. The aim of this study was to automate cardiovascular magnetic resonance (CMR) RA area measurements and evaluate the clinical utility by assessing repeatability, correlation with invasive haemodynamics and prognostic value.

METHODS

A deep learning RA area CMR contouring model was trained in a multicentre cohort of 365 patients with pulmonary hypertension, left ventricular pathology and healthy subjects. Inter-study repeatability (intraclass correlation coefficient (ICC)) and agreement of contours (DICE similarity coefficient (DSC)) were assessed in a prospective cohort (n = 36). Clinical testing and mortality prediction was performed in n = 400 patients that were not used in the training nor prospective cohort, and the correlation of automatic and manual RA measurements with invasive haemodynamics assessed in n = 212/400. Radiologist quality control (QC) was performed in the ASPIRE registry, n = 3795 patients. The primary QC observer evaluated all the segmentations and recorded them as satisfactory, suboptimal or failure. A second QC observer analysed a random subcohort to assess QC agreement (n = 1018).

RESULTS

All deep learning RA measurements showed higher interstudy repeatability (ICC 0.91 to 0.95) compared to manual RA measurements (1st observer ICC 0.82 to 0.88, 2nd observer ICC 0.88 to 0.91). DSC showed high agreement comparing automatic artificial intelligence and manual CMR readers. Maximal RA area mean and standard deviation (SD) DSC metric for observer 1 vs observer 2, automatic measurements vs observer 1 and automatic measurements vs observer 2 is 92.4 ± 3.5 cm2, 91.2 ± 4.5 cm2 and 93.2 ± 3.2 cm2, respectively. Minimal RA area mean and SD DSC metric for observer 1 vs observer 2, automatic measurements vs observer 1 and automatic measurements vs observer 2 was 89.8 ± 3.9 cm2, 87.0 ± 5.8 cm2 and 91.8 ± 4.8 cm2. Automatic RA area measurements all showed moderate correlation with invasive parameters (r = 0.45 to 0.66), manual (r = 0.36 to 0.57). Maximal RA area could accurately predict elevated mean RA pressure low and high-risk thresholds (area under the receiver operating characteristic curve artificial intelligence = 0.82/0.87 vs manual = 0.78/0.83), and predicted mortality similar to manual measurements, both p < 0.01. In the QC evaluation, artificial intelligence segmentations were suboptimal at 108/3795 and a low failure rate of 16/3795. In a subcohort (n = 1018), agreement by two QC observers was excellent, kappa 0.84.

CONCLUSION

Automatic artificial intelligence CMR derived RA size and function are accurate, have excellent repeatability, moderate associations with invasive haemodynamics and predict mortality.

High interstudy repeatability of automatic deep learnt biventricular CMR measurements

Funding Acknowledgements

Type of funding sources: Foundation. Main funding source(s): Wellcome Trust (UK), NIHR (UK)

Introduction

Cardiac magnetic resonance (CMR) measurements have significant diagnostic and prognostic value. Accurate and repeatable measurements are essential to assess disease severity, evaluate therapy response and monitor disease progression. Deep learning approaches have shown promise for automatic left ventricular (LV) segmentation on CMR, however fully automatic right ventricular (RV) segmentation remains challenging. We aimed to develop a biventricular automatic contouring model and evaluate the interstudy repeatability of the model in a prospectively recruited cohort.

Methods

A deep learning CMR contouring model was developed in a retrospective multi-vendor (Siemens and General Electric), multi-pathology cohort of patients, predominantly with heart failure, pulmonary hypertension and lung diseases (n = 400, ASPIRE registry). Biventricular segmentations were made on all CMR studies across cardiac phases. To test the accuracy of the automatic segmentation, 30 ASPIRE CMRs were segmented independently by two CMR experts. Each segmentation was compared to the automatic contouring with agreement assessed using the Dice similarity coefficient (DSC). 

A prospective validation cohort of 46 subjects (10 healthy volunteers and 36 patients with pulmonary hypertension) were recruited to assess interstudy agreement of automatic and manual CMR assessments. Two CMR studies were performed during separate sessions on the same day. Interstudy repeatability was assessed using intraclass correlation coefficient (ICC) and Bland-Altman plots. 

Results

DSC showed high agreement (figure 1) comparing automatic and expert CMR readers, with minimal bias towards either CMR expert. The scan-scan repeatability CMR measurements were higher for all automatic RV measurements (ICC 0.89 to 0.98) compared to manual RV measurements (0.78 to 0.98). LV automatic and manual measurements were similarly repeatable (figure 2). Bland-Altman plots showed strong agreement with small mean differences between the scan-scan measurements (figure 2).

Conclusion

Fully automatic biventricular short-axis segmentations are comparable with expert manual segmentations, and have shown excellent interstudy repeatability.

Fully automated CMR derived stroke volume correlates with right heart catheter measurements in patients with suspected pulmonary hypertension

Funding Acknowledgements

Type of funding sources: Foundation. Main funding source(s): Welcome Trust (UK), NIHR (UK)

Introduction

Cardiac magnetic resonance (CMR) assessment plays a significant role in the diagnosis, prognosis and monitoring of patients with pulmonary hypertension (PH). We developed a deep learning model to automatically generate biventricular contours and validated its result in a prospective cohort of patients with suspected PH who underwent right heart catheterization (RHC).

Methods

A deep learning CMR contouring model was developed in a retrospective multi-vendor (Siemens and General Electric), multi-pathology cohort of patients, predominantly with heart failure, lung disease and PH (n = 400, ASPIRE registry). Biventricular segmentations were made on all CMR studies across cardiac phases. A prospective validation cohort of 102 suspected PH patients was recruited and they had RHC within 24 hours  of the CMR. To test the accuracy of the automatic segmentation, the RHC-thermodilution and CMR-derived measures of stroke volume (SV) were compared for manual and automated measurements. 

Results

The mean and standard deviation for the derived SV was 59 ml ± 21 measured by RHC and 75 ml ± 25 for automated and 79 ml ± 26 for manual CMR measurements. Automatic and manual CMR measurement correlated strongly with RHC derived SV; 0.73, 95% CI [0.62, 0.81] and 0.78, 95% CI [0.69, 0.85], respectively (figure 1).  The agreement between automatic and manual SV was high; interclass correlation coefficient (ICC) = 0.88, 95% CI [0.83, 0.92] and Bland-Altman plots showed a narrow spread of mean differences between manual and automatic measurements (figure 2).

Conclusion

In a prospective cohort, fully automatic CMR assessments corresponded accurately to invasive hemodynamics performed within 24 hours of a CMR study.

Can we optimize electrode placement for impedance pneumography?

In this paper, we discuss issues involved in defining an optimum placement of four electrodes for impedance pneumography. We observed a general trend where the change in impedance (delta Z) decreased while the sensitivity (delta Z/Z) increased with distance between the drive and receive electrode pairs. However, the theoretical study indicated that delta Z/Z should decrease with distance. The scatter of points in the plots indicated that sensitivity was influenced by factors other than distance. The correlation coefficient between the theoretical and measured delta Z/Z was low, but significant. This suggested that the best electrode configuration can be derived from the theoretical data. High delta Z/Z was obtained when the drive and receive electrode pairs were placed close to the lungs and in different horizontal planes.

Three-dimensional electrical impedance tomography

The electrical resistivity of mammalian tissues varies widely and is correlated with physiological function. Electrical impedance tomography (EIT) can be used to probe such variations in vivo, and offers a non-invasive means of imaging the internal conductivity distribution of the human body. But the computational complexity of EIT has severe practical limitations, and previous work has been restricted to considering image reconstruction as an essentially two-dimensional problem. This simplification can limit significantly the imaging capabilities of EIT, as the electric currents used to determine the conductivity variations will not in general be confined to a two-dimensional plane. A few studies have attempted three-dimensional EIT image reconstruction, but have not yet succeeded in generating images of a quality suitable for clinical applications. Here we report the development of a three-dimensional EIT system with greatly improved imaging capabilities, which combines our 64-electrode data-collection apparatus with customized matrix inversion techniques. Our results demonstrate the practical potential of EIT for clinical applications, such as lung or brain imaging and diagnostic screening.

Posture control using electrical stimulation biofeedback: a pilot study

The investigation studied the effects of biofeedback on the sitting posture of a 14 year old girl with cerebral palsy. The subject's posture was quantified using a video analysis technique which established the threshold of poor posture at 30 degrees from the vertical plane. A stimulator system was designed using an adapted drop foot stimulator and a custom made controller with a mercury tilt switch as the posture angle transducer. If posture became greater than 30 degrees tactile electrical stimulation was administered to the subject's lower back. Repetitive stimuli occurred on non-correction of posture, with a maximum of 4 consecutive stimuli, upon which an alarm was activated. 10 training sessions of 20 min duration were completed over a 4 week period, monitored using a data logger. Following initial improvement the daily results show a gradual deterioration in posture, whilst post-trial video analysis indicates a significant improvement in posture. An improved response to the alarm stimulus is observed. Reasons for these conflicting findings are discussed.