Weina Jin

Clinical evaluation evidence and model explainability are key gatekeepers to ensure the safe, accountable, and effective use of artificial intelligence (AI) in clinical settings. We conducted a clinical user-centered evaluation with 35 neurosurgeons to assess the utility of AI assistance and its explanation on the glioma grading task. Each participant read 25 brain MRI scans of patients with gliomas, and gave their judgment on the glioma grading without and with the assistance of AI prediction and explanation. The AI model was trained on the BraTS dataset with 88.0% accuracy. The AI explanation was generated using the explainable AI algorithm of SmoothGrad, which was selected from 16 algorithms based on the criterion of being truthful to the AI decision process. Results showed that compared to the average accuracy of 82.5±8.7% when physicians performed the task alone, physicians’ task performance increased to 87.7±7.3% with statistical significance (p-value = 0.002) when assisted by AI prediction, and remained at almost the same level of 88.5±7.0% (p-value = 0.35) with the additional assistance of AI explanation. Based on quantitative and qualitative results, the observed improvement in physicians’ task performance assisted by AI prediction was mainly because physicians’ decision patterns converged to be similar to AI, as physicians only switched their decisions when disagreeing with AI. The insignificant change in physicians’ performance with the additional assistance of AI explanation was because the AI explanations did not provide explicit reasons, contexts, or descriptions of clinical features to help doctors discern potentially incorrect AI predictions. The evaluation showed the clinical utility of AI to assist physicians on the glioma grading task, and identified the limitations and clinical usage gaps of existing explainable AI techniques for future improvement.

Explainable artificial intelligence (XAI) is essential for enabling clinical users to get informed decision support from AI and comply with evidence-based medical practice. Applying XAI in clinical settings requires proper evaluation criteria to ensure the explanation technique is both technically sound and clinically useful, but specific support is lacking to achieve this goal. To bridge the research gap, we propose the Clinical XAI Guidelines that consist of five criteria a clinical XAI needs to be optimized for. The guidelines recommend choosing an explanation form based on Guideline 1 (G1) Understandability and G2 Clinical relevance. For the chosen explanation form, its specific XAI technique should be optimized for G3 Truthfulness, G4 Informative plausibility, and G5 Computational efficiency. Following the guidelines, we conducted a systematic evaluation on a novel problem of multi-modal medical image explanation with two clinical tasks, and proposed new evaluation metrics accordingly. Sixteen commonly-used heatmap XAI techniques were evaluated and found to be insufficient for clinical use due to their failure in G3 and G4. Our evaluation demonstrated the use of Clinical XAI Guidelines to support the design and evaluation of clinically viable XAI.

Being able to explain the prediction to clinical end-users is a necessity to leverage the power of artificial intelligence (AI) models for clinical decision support. For medical images, a feature attribution map, or heatmap, is the most common form of explanation that highlights important features for AI models’ prediction. However, it is unknown how well heatmaps perform on explaining decisions on multi-modal medical images, where each image modality or channel visualizes distinct clinical information of the same underlying biomedical phenomenon. Understanding such modality-dependent features is essential for clinical users’ interpretation of AI decisions. To tackle this clinically important but technically ignored problem, we propose the modality-specific feature importance (MSFI) metric. It encodes clinical image and explanation interpretation patterns of modality prioritization and modality-specific feature localization. We conduct a clinical requirement-grounded, systematic evaluation using computational methods and a clinician user study. Results show that the examined 16 heatmap algorithms failed to fulfill clinical requirements to correctly indicate AI model decision process or decision quality. The evaluation and MSFI metric can guide the design and selection of explainable AI algorithms to meet clinical requirements on multi-modal explanation.