Personalized artificial intelligence based left ventricular ejection fraction and systolic dysfunction assessment

Figures

• 

  Fig. 1. Sex-based LVEF distribution of the TTE cohort. Histogram showing the distribution of left ventricular ejection fraction (LVEF) in the study cohort stratified by sex. Male subjects are shown in blue and female subjects in red.

• 

  Fig. 2. Error distribution plot of the ECG-only model of CNN architecture. The plot shows the distribution of errors within each range, MAE and sample counts for each range are annotated above the plot.

• 

  Fig. 3. Error distribution plot of the hybrid model of CNN architecture. The plot shows the distribution of errors within each range, the MAE and sample counts for each range are annotated above the plot.

• 

  Fig. 4. Error distribution plot of the ECG-only personalized CNN model architecture. The box plots show the distribution of relative error for each sample number across subjects, with the upper whisker indicating the 95th percentile (dark blue line). Red crosses represent individual subject-level errors that exceed this threshold, corresponding to the top 5% of relative errors for each sample number. A fitted line is used to highlight the trend of decreasing uncertainty with every additional new sample. The number of subjects contributing to each sample number is presented above the plot.

• 

  Fig. 5. Comparison of the error distribution between the personalized and generalized CNN models using ECG-only data. The plot shows the distribution of errors across each LVEF range.

• 

  Fig. 6. Error distribution plot using the hybrid CNN personalized model. The box plots show the distribution of relative error for each sample number across subjects, with the upper whisker indicating the 95th percentile (dark blue line). Red crosses represent individual subject-level errors that exceed this threshold, corresponding to the top 5% of relative errors for each sample number. A fitted line is used to highlight the trend of decreasing uncertainty with every additional new sample. The number of subjects contributing to each sample number is presented above the plot.

• 

  Fig. 7. Comparison of absolute error distributions between the personalized and generalized CNN models using the hybrid CNN architecture. The plot illustrates error distributions across different LVEF ranges.