Table 3 Benchmarking with state-of-the-art CNN classifier models
From: Automatic echocardiographic anomalies interpretation using a stacked residual-dense network model
Algorithm | Period | Number of class abnormality | Performance (%) | ||
|---|---|---|---|---|---|
Accuracy | Sensitivity | Specificity | |||
Residual learning [15] | Prenatal | 2 classes (normal vs diseased) validation data | 93 | 93 | – |
2 classes (normal vs diseased) unseen data | 91 | 91 | – | ||
Deep learning model [23] | Prenatal | 2 classes (normal vs TOF validation data | – | 75 | 76 |
2 classes (normal vs HLHS) validation data | – | 100 | 90 | ||
DGACNN [24] | Prenatal | 2 classes (normal and diseased) validation data | 85 | – | – |
Proposed Stacked model | Incorporating prenatal and postnatal | 4 classes (normal, ASD, VSD, AVSD) validation data | 99 | 99 | 99 |
4 classes (normal, ASD, VSD, AVSD) unseen data | 92 | 92 | 94 | ||