Perception-Aligned AI Outputs: End-to-End Visual Prediction for Uncertainty Communication in Clinical Decision-Making

Visual explanations outperform traditional numerical outputs for clinical decision-making: The study found that 84% of participants found visual explanations more memorable than numeric outputs, and 76.9% reported faster decision-making when using VL4ML compared to conventional explanation methods.

Non-technical audiences can understand AI uncertainty without statistical training: Over 82% of study participants successfully perceived uncertainty embedded in visual representations without prior statistical knowledge, suggesting that well-designed visual communication can make complex AI concepts accessible to diverse audiences including patients and non-specialist clinicians.

Visual explainability frameworks promote equitable AI understanding across different groups: The research found no significant differences in comprehension between clinicians and non-clinicians or between male and female participants, indicating that perception-aligned visual explanations can democratize AI literacy rather than creating knowledge gaps based on professional background or demographics.

Human-centered design principles improve trust in clinical AI systems: With 79% of the 158 study participants (including 39.2% clinical professionals) positively rating the visual explanations across multiple evaluation dimensions, the findings demonstrate that prioritizing how humans perceive and interpret AI outputs—rather than just technical accuracy—is crucial for building trustworthy healthcare AI.