Table 4 Expert ratings of SHAP-based multimodal explanations.
From: Explainable federated transformer framework for joint leukemia classification and stage prediction
Evaluation criterion | Mean score (1–5) | Standard deviation |
|---|---|---|
Relevance of visual SHAP highlights | 4.6 | 0.41 |
Relevance of text token attributions | 4.4 | 0.52 |
Alignment with clinical diagnostic reasoning | 4.5 | 0.47 |
Usefulness in verifying model predictions | 4.7 | 0.38 |
Helpfulness in ambiguous or borderline cases | 4.3 | 0.55 |
Trust enhancement due to explanations | 4.5 | 0.49 |
Perceived readiness for clinical deployment (explainability module) | 4.2 | 0.59 |
