Figure 5 | Scientific Reports

Figure 5

From: A deep learning framework assisted echocardiography with diagnosis, lesion localization, phenogrouping heterogeneous disease, and anomaly detection

Figure 5

Distinct phenogroups identification. (a,b) The top row presented six cases with raw echocardiogram images of the DCM-high patients (1–3) and the DCM-low patients (4–6), corresponding with CAM heatmaps in the bottom row. (c) PHATE (Potential of Heat-diffusion for Affinity-based Trajectory Embedding) was used as a dimensionality reduction method to visualize structures projected from a 2048-dimensional array generated by the last hidden layer of the Inception-V3 model into a two-dimensional array (PHATE 1 and 2). (d) P-values of 63 clinical variables in total, sorting from the lowest value to the highest. P < 0.05, with a statistical difference, was defined as the threshold plotted in a blue dashed line; P < 0.000794 with a statistical difference using the Bonferroni method, was defined as the threshold plotted in a light blue dashed line. Traits commonly related to cardiac conditions are labeled in red, while others are labeled in green. (e–n) Histograms of 10 clinical characteristics with significant statistical differences using the Bonferroni method.

Back to article page