Echo2Pheno: a deep-learning application to uncover echocardiographic phenotypes in conscious mice

Abstract

Echocardiography, a rapid and cost-effective imaging technique, assesses cardiac function and structure. Despite its popularity in cardiovascular medicine and clinical research, image-derived phenotypic measurements are manually performed, requiring expert knowledge and training. Notwithstanding great progress in deep-learning applications in small animal echocardiography, the focus has so far only been on images of anesthetized rodents. We present here a new algorithm specifically designed for echocardiograms acquired in conscious mice called Echo2Pheno, an automatic statistical learning workflow for analyzing and interpreting high-throughput non-anesthetized transthoracic murine echocardiographic images in the presence of genetic knockouts. Echo2Pheno comprises a neural network module for echocardiographic image analysis and phenotypic measurements, including a statistical hypothesis-testing framework for assessing phenotypic differences between populations. Using 2159 images of 16 different knockout mouse strains of the German Mouse Clinic, Echo2Pheno accurately confirms known cardiovascular genotype-phenotype relationships (e.g., Dystrophin) and discovers novel genes (e.g., CCR4-NOT transcription complex subunit 6-like, Cnot6l, and synaptotagmin-like protein 4, Sytl4), which cause altered cardiovascular phenotypes, as verified by H&E-stained histological images. Echo2Pheno provides an important step toward automatic end-to-end learning for linking echocardiographic readouts to cardiovascular phenotypes of interest in conscious mice.

Fig. 1

Manual versus automatic annotations of LVIDd and LVIDs in the 16 validation studies. a Scatter plot of LVIDx_auto and LVIDx_man and Pearson correlation coefficients (r_d $=0.69$ and r_s $=0.84$). b Box plots of the difference LVIDx_auto-LVIDx_man

Fig. 2

Phenotype detection with Echo2Pheno compared to the manual analysis in the 16 validation studies. Wilcoxon-Rank-Sum-Test performed  p-values for left ventricular inner diameter in systole (LVIDs), left ventricular inner diameter in diastole (LVIDd), ejection fraction (EF), and fractional shortening (FS) in male and female mice and classified the studies in Panel a Validation of manually identified phenotypes; Panel b Loss of manually identified phenotypes and Panel c Detection of manually unidentified phenotypes.

Fig. 3

Representative echocardiograms (Visual Sonics, Vevo2100, short axis, M-mode) for comparison of manually and automatically extracted features. Manual annotation shown in the left panel is based on only two consecutive heartbeats whereas automatic Echo2Pheno-based annotation is considering the entire dataset. Panel a Validation of manually identified phenotypes by an agreement of manual (left panel) and automatic (right panel) analysis in Gstm1; Panel b Loss of manually identified phenotypes in LVIDs, EF, and FS due to a mismatch between manual (left) and automatic (right) analysis in Uggt2; Panel c Detection of manually unidentified phenotypes in Cnot6l (LVIDs for both sexes) not found with the manual annotations

Fig. 4

Histopathological evaluation in the 16 validation studies. Representative photomicrographs of hematoxylin and eosin-stained, paraffin-embedded longitudinal heart sections illustrating left (LV) and right (RV) ventricles. Panel a shows a representative example of a structurally healthy heart with normal LV dimensions in a male (left) and female (right) C57BL/6N mouse, followed by all 16 mouse validation studies Panels b–d. Studies color coded in orange show LV dilation

Fig. 5

Panel a representative echocardiogram (Visual Sonics, Vevo2100 system) illustrating data quality labeled with color across the DICOM file. Below, yellow indicates good quality traces at a cut-off of 0.5, blue indicates low quality (b-c). Panel b illustrates the parameter profile of end-diastolic left ventricular inner diameter (LVIDd) and Panel c the profile of end-systolic left ventricular inner diameter (LVIDs) over recording time (ms) in one representative male Acnat mutant mouse

Fig. 6

Image quality per study. The quality of the 16 different studies is presented in percentages of high-quality (yellow) and poor-quality (blue) regions