Fig 1.
Overall architecture of our method.
Table 1.
Notation and definitions for important variables.
Fig 2.
Quantitative evaluation of our proposed method on ACDC and MMWHS datasets under various labeled data ratios.
Table 2.
Comparison with state-of-the-art methods on ACDC and MMWHS datasets under different labeled data ratios.
Bold: best result. Underline: second best result.
Table 3.
Detailed comparison with state-of-the-art methods across various metrics on ACDC with a labeled data ratio of 10%.
Fig 3.
Qualitative comparison of different methods on the ACDC and MMWHS datasets.
For ACDC, we use 2.5% labeled data. For MMWHS, we use 20% labeled data.
Fig 4.
Visualization of dynamic pseudo-label threshold maps of an example in the ACDC dataset.
c represents the class.
Table 4.
Per-structure segmentation results on the ACDC dataset (2.5% labeled data).
Results are reported as mean ± standard deviation over 3 runs. RV: Right Ventricle, Myo: Myocardium, LV: Left Ventricle.
Table 5.
Comparison of performance between our dynamic pseudo-label threshold map (DPTM) strategy with recent pseudo-label-based methods.
For a fair comparison, we only implement the pseudo-label part in these methods (denoted as †) and then use the pseudo-labels for self-training. We report the average DSC in the table. Bold: best result. Underline: second best result.
Fig 5.
Grad-CAM visualization of the proposed method on the ACDC and MMWHS datasets.
The top row shows the failure results on the ACDC dataset and the bottom row shows the successful results on the MMWHS dataset.
Table 6.
Benefits of our proposed module dynamic pseudo-label threshold map (DPTM), robust entropy minimization (REM), and contrastive consistency (CC).
For ACDC, we use 10% labeled data. For MMWHS, we use 40% labeled data. † denotes the best performance when attached with only one component and * represents the best results attached with two components.
Fig 6.
Benefits of the three proposed components dynamic pseudo-label threshold map (DPTM), robust entropy minimization (REM) and contrastive consistency (CC).
Fig 7.
The impact of on the performance of the model on two datasets.
Fig 8.
The impact of on the performance of the model on two datasets.
Table 7.
Computational efficiency comparison on ACDC dataset (10% labeled data).
Table 8.
Transfer learning comparison from CHD to MMWHS.
All methods are pre-trained on CHD and fine-tuned on MMWHS with varying numbers of labeled patients (M).