Table 5 Summary of implementation details and hyperparameters.
Category | Hyperparameter | Value/method |
|---|---|---|
Data & Model | Input Image Size | 112 × 112 pixels |
Data Split Strategy | Pre-defined splits provided with each dataset were used. EchoNet-LVH: Train: 8000/Val: 1000/Test: 1030 EchoNet-Dynamic: Train: 7465/Val: 1064/Test: 1024 CAMUS: Train: 450/Test: 150 | |
Loss Function | Segmentation: Dice Loss + Binary Cross-Entropy Keypoint: MSE on heatmaps | |
Training | Optimizer | Adam |
Learning Rate | 0.001 | |
Number of Epochs | 20 | |
Batch Size | Based on GPU memory capacity | |
Optimization | Hyperparameter Tuning | Automated using the Optuna framework (20 trials) |
Loss Weighting Factor (alpha) | Determined via Optuna tuning to balance task losses. | |
System | Operating System | Linux-based x64 |
CPU | Intel® Core™ i9 | |
GPU | NVIDIA GeForce RTX 3080 Ti | |
RAM | 64 GB |
