Table 2.
Statistics for each algorithm.
| Group N | Group A | Group B | Group C | ||
|---|---|---|---|---|---|
| Conventional | PPV | – | 0.81 | 0.29 | 0.40 |
| Sensitivity | – | 0.71 | 0.39 | 0.46 | |
| F1-score | – | 0.76 | 0.33 | 0.43 | |
| Accuracy | 0.61 | ||||
| ECG | PPV, mean (SD) | 0.78 (0.04) | 0.90 (0.01) | 0.56 (0.05) | 0.49 (0.02) |
| p value* | – | < 0.001 | < 0.001 | < 0.001 | |
| Sensitivity, mean (SD) | 0.97 (0.01) | 0.86 (0.02) | 0.61 (0.04) | 0.40 (0.04) | |
| p value* | – | < 0.001 | < 0.001 | 0.84 | |
| F1-score, mean (SD) | 0.86 (0.02) | 0.88 (0.02) | 0.58 (0.04) | 0.44 (0.03) | |
| p value* | – | < 0.001 | < 0.001 | 0.019 | |
| Accuracy, mean (SD) | 0.78 (0.02) | ||||
| p value* | < 0.001 | ||||
| ECG and X-ray | PPV, mean (SD) | 0.80 (0.04) | 0.91 (0.02) | 0.63 (0.05) | 0.58 (0.06) |
| p value+ | 0.44 | 0.23 | 0.09 | 0.06 | |
| Sensitivity, mean (SD) | 0.99 (0.02) | 0.88 (0.02) | 0.73 (0.04) | 0.40 (0.07) | |
| p value+ | 0.10 | 0.22 | 0.0011 | 0.89 | |
| F1-score, mean (SD) | 0.88 (0.03) | 0.89 (0.01) | 0.67 (0.05) | 0.46 (0.07) | |
| p value+ | 0.23 | 0.24 | 0.008 | 0.64 | |
| Accuracy, mean (SD) | 0.80 (0.02) | ||||
| p value+ | 0.040 | ||||
We analysed the performance of each model. The conventional model was performed once, and the ‘ECG’ and ‘ECG and X-ray’ model was repeatedly performed to analyse the statistics. PPV indicates the positive predictive value; ECG is the abbreviation for echocardiogram. Data are expressed as the mean (standard deviation). The *p value compares the statistics of the conventional algorithm and the deep learning model using only ECG data. The +p value compares the statistics of the deep learning model using only ECGs and the model using both ECG and chest X-ray data.