Background: Heart failure with preserved ejection fraction (HFpEF) is a heterogeneous syndrome requiring invasive and/or exercise stress testing for diagnosis, in contrast with the homogeneous nature of HF with reduced EF (HFrEF). We previously demonstrated that acoustic cardiography (AC), combined with advanced signal analysis, can non-invasively estimate left ventricular (LV) functional indices in an experimental swine model of HFrEF. Objectives: To evaluate whether AC-derived features capture stress-dependent LV functional changes, supporting phenotypic differentiation between porcine models of HFpEF, HFrEF, and Control conditions. Methods: Synchronized invasive LV-pressure and non-invasive ECG, pulse oximetry, and AC signals were collected from 12 anesthetized, closed-chest Göttingen minipigs (Control, n = 3; HFrEF, n = 5; HFpEF, n = 4). Through signal analysis, we derived time and frequency features from the non-invasive signals to predict, using our AI model, the invasively measured LV functional indices. Atrial pacing was performed up to 160 bpm as a controlled heart-rate stress paradigm. Two AI modeling strategies were employed: a standard 80/20 train/test ratio, and a leave-one-animal-out (i.e., 1 animal per health status) to assess generalizability. Results: Standard blind testing achieved >95% accuracy in phenotype classification with <3% relative error for predicted LV indices. The leave-one-animal-out classification performance remained robust (79–95% accuracy), supporting translational potential despite inter-animal variability. Notably, HFpEF animals exhibited greater variability in AC-features across increasing heart rates compared to Control and HFrEF. Conclusions: AC-based modelling offers a rapid, non-invasive approach for assessing LV function. Our methodology may complement existing diagnostic tools, particularly for conditions like HFpEF, but warrants further validation in clinical populations.
Acoustic Cardiography Captures Stress-Dependent Ventricular Dysfunction and Distinguishes Heart Failure Phenotypes in a Porcine Model / Lo Muzio, F.P., Fassina, L., Otvos, J., Wierling, K., Baer, S., Berboth, L., Kind, A., Faragli, A., Alogna, A.. - In: BIOMEDICINES. - ISSN 2227-9059. - 14:8(2026). [10.3390/biomedicines14081669]
Acoustic Cardiography Captures Stress-Dependent Ventricular Dysfunction and Distinguishes Heart Failure Phenotypes in a Porcine Model
Lo Muzio F. P.;
2026-01-01
Abstract
Background: Heart failure with preserved ejection fraction (HFpEF) is a heterogeneous syndrome requiring invasive and/or exercise stress testing for diagnosis, in contrast with the homogeneous nature of HF with reduced EF (HFrEF). We previously demonstrated that acoustic cardiography (AC), combined with advanced signal analysis, can non-invasively estimate left ventricular (LV) functional indices in an experimental swine model of HFrEF. Objectives: To evaluate whether AC-derived features capture stress-dependent LV functional changes, supporting phenotypic differentiation between porcine models of HFpEF, HFrEF, and Control conditions. Methods: Synchronized invasive LV-pressure and non-invasive ECG, pulse oximetry, and AC signals were collected from 12 anesthetized, closed-chest Göttingen minipigs (Control, n = 3; HFrEF, n = 5; HFpEF, n = 4). Through signal analysis, we derived time and frequency features from the non-invasive signals to predict, using our AI model, the invasively measured LV functional indices. Atrial pacing was performed up to 160 bpm as a controlled heart-rate stress paradigm. Two AI modeling strategies were employed: a standard 80/20 train/test ratio, and a leave-one-animal-out (i.e., 1 animal per health status) to assess generalizability. Results: Standard blind testing achieved >95% accuracy in phenotype classification with <3% relative error for predicted LV indices. The leave-one-animal-out classification performance remained robust (79–95% accuracy), supporting translational potential despite inter-animal variability. Notably, HFpEF animals exhibited greater variability in AC-features across increasing heart rates compared to Control and HFrEF. Conclusions: AC-based modelling offers a rapid, non-invasive approach for assessing LV function. Our methodology may complement existing diagnostic tools, particularly for conditions like HFpEF, but warrants further validation in clinical populations.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
