Explainability Through Human-Centric Design for XAI in Lung Cancer Detection

Human-centered design improves clinical trustworthiness: XpertXAI was specifically built around how radiologists actually think about lung cancer diagnosis, rather than applying generic explanation techniques after the fact. This expert-guided approach resulted in explanations that better aligned with radiologist judgments compared to leading post-hoc explainability methods.

Standard AI explanation methods fall short in medicine: The paper found that existing explainability techniques "frequently fail to produce clinically meaningful explanations, omitting key diagnostic features and disagreeing with radiologist judgments"—highlighting that off-the-shelf AI explanation tools cannot simply be transplanted into healthcare without domain expertise.

Concept-based explanations bridge the AI-clinician gap: By preserving human-interpretable clinical concepts rather than relying on abstract model features, XpertXAI enables doctors to understand why the AI reached its diagnosis in terms they recognize and can validate, addressing a major barrier to clinical adoption of AI systems.

Scalability and accuracy can coexist with explainability: XpertXAI achieved both superior predictive accuracy and better explanations while being designed to detect multiple lung pathologies, suggesting that interpretability doesn't require sacrificing performance or limiting AI systems to single diagnostic tasks.

Expert validation is essential for responsible AI deployment: The researchers validated their explanations against actual radiologist annotations and medical ground truth, demonstrating that responsible AI in healthcare requires rigorous comparison with human expert judgment rather than relying solely on algorithmic metrics.