Population data-based federated machine learning improves automated echocardiographic quantification of cardiac structure and function: the Automatisierte Vermessung der Echokardiographie project.

Aims: Machine-learning (ML)-based automated measurement of echocardiography images emerges as an option to reduce observer variability. The objective of the study is to improve the accuracy of a pre-existing automated reading tool ('original detector') by federated ML-based re-training. Methods and results: Automatisierte Vermessung der Echokardiographie was based on the echocardiography images of n = 4965 participants of the population-based Characteristics and Course of Heart Failure Stages A-B and Determinants of Progression Cohort Study. We implemented federated ML: echocardiography images were read by the Academic Core Lab Ultrasound-based Cardiovascular Imaging at the University Hospital Würzburg (UKW). A random algorithm selected 3226 participants for re-training of the original detector. According to data protection rules, the generation of ground truth and ML training cycles took place within the UKW network. Only non-personal training weights were exchanged with the external cooperation partner for the refinement of ML algorithms. Both the original detectors as the re-trained detector were then applied to the echocardiograms of n = 563 participants not used for training. With regard to the human referent, the re-trained detector revealed (i) superior accuracy when contrasted with the original detector's performance as it arrived at significantly smaller mean differences in all but one parameter, and a (ii) smaller absolute difference between measurements when compared with a group of different human observers. Conclusion: Population data-based ML in a federated ML set-up was feasible. The re-trained detector exhibited a much lower measurement variability than human readers. This gain in accuracy and precision strengthens the confidence in automated echocardiographic readings, which carries large potential for applications in various settings.

Quantitative analysis of left ventricular strain using cardiac computed tomography.

OBJECTIVES: To investigate whether cardiac computed tomography (CCT) can determine left ventricular (LV) radial, circumferential and longitudinal myocardial deformation in comparison to two-dimensional echocardiography in patients with congestive heart failure. BACKGROUND: Echocardiography allows for accurate assessment of strain with high temporal resolution. A reduced strain is associated with a poor prognosis in cardiomyopathies. However, strain imaging is limited in patients with poor echogenic windows, so that, in selected cases, tomographic imaging techniques may be preferable for the evaluation of myocardial deformation. METHODS: Consecutive patients (n=27) with congestive heart failure who underwent a clinically indicated ECG-gated contrast-enhanced 64-slice dual-source CCT for the evaluation of the cardiac veins prior to cardiac resynchronization therapy (CRT) were included. All patients underwent additional echocardiography. LV radial, circumferential and longitudinal strain and strain rates were analyzed in identical midventricular short axis, 4-, 2- and 3-chamber views for both modalities using the same prototype software algorithm (feature tracking). Time for analysis was assessed for both modalities. RESULTS: Close correlations were observed for both techniques regarding global strain (r=0.93, r=0.87 and r=0.84 for radial, circumferential and longitudinal strain, respectively, p<0.001 for all). Similar trends were observed for regional radial, longitudinal and circumferential strain (r=0.88, r=0.84 and r=0.94, respectively, p<0.001 for all). The number of non-diagnostic myocardial segments was significantly higher with echocardiography than with CCT (9.6% versus 1.9%, p<0.001). In addition, the required time for complete quantitative strain analysis was significantly shorter for CCT compared to echocardiography (877±119 s per patient versus 1105±258 s per patient, p<0.001). CONCLUSION: Quantitative assessment of LV strain is feasible using CCT. This technique may represent a valuable alternative for the assessment of myocardial deformation in selected patients with poor echogenic windows and general contraindications for magnetic resonance imaging.

Beat to beat 3-dimensional intracardiac echocardiography: theoretical approach and practical experiences.

Three-dimensional (3D)-imaging provides important information on cardiac anatomy during electrophysiological procedures. Real-time updates of modalities with high soft-tissue contrast are particularly advantageous during cardiac procedures. Therefore, a beat to beat 3D visualization of cardiac anatomy by intracardiac echocardiography (ICE) was developed and tested in phantoms and animals. An electronic phased-array 5-10 MHz ICE-catheter (Acuson, AcuNav/Siemens Medical Solutions USA/64 elements) providing a 90° sector image was used for ICE-imaging. A custom-made mechanical prototype controlled by a servo motor allowed automatic rotation of the ICE-catheter around its longitudinal axis. During a single heartbeat, the ICE-catheter was rotated and 2D-images were acquired. Reconstruction into a 3D volume and rendering by a prototype software was performed beat to beat. After experimental validation using a rigid phantom, the system was tested in an animal study and afterwards, for quantitative validation, in a dynamic phantom. Acquisition of beat to beat 3D-reconstruction was technically feasible. However, twisting of the ICE-catheter shaft due to friction and torsion was found and rotation was hampered. Also, depiction of catheters was not always ensured in case of parallel alignment. Using a curved sheath for depiction of cardiac anatomy there was no congruent depiction of shape and dimension of static and moving objects. Beat to beat 3D-ICE-imaging is feasible. However, shape and dimension of static and moving objects cannot always be displayed with necessary steadiness as needed in the clinical setting. As catheter depiction is also limited, clinical use seems impossible.

Interactive image processing as learning by doing component of a WBT application.

Web Based Training (WBT) is of increasing importance [1][2]. In [5] the main author has presented and discussed an overall approach to WBT for medical image processing. In the present contribution the interactive part of the approach is discussed in detail and its prototypical realisation in the framework of a client server teaching application is presented.