Table 2 Instance segmentation performance with two backbones: ResNet50 and ResNet101 architectures
From: Automatic echocardiographic anomalies interpretation using a stacked residual-dense network model
Condition | Validation data | Unseen data | ||
|---|---|---|---|---|
ResNet50 | ResNet101 | ResNet50 | ResNet101 | |
IoU (%) | ||||
ASD | 55.69 | 57.07 | 26.96 | 41.21 |
AVSD | 65.23 | 60.38 | 16.75 | 52.08 |
VSD | 53.90 | 57.07 | 22.43 | 33.89 |
DSC (%) | ||||
ASD | 72.52 | 77.03 | 40.84 | 49.12 |
AVSD | 75.82 | 74.39 | 44.20 | 73.97 |
VSD | 75.17 | 75.84 | 46.84 | 44.06 |
mAP (%) | ||||
Average | 73.92 | 76.36 | 31.11 | 51.04 |