Diagnosing rare diseases remains a persistent challenge, often hindered by cognitive anchoring: once clinicians settle on a common diagnosis, alternative-especially rare-explanations are often dismissed. To address this, we propose a human - centered counterfactual reasoning framework using a Denoising Autoencoder (DAE) to simulate what - if diagnostic scenarios that disrupt clinicians’ initial assumptions. Our model uniquely and jointly learns (1) the true distribution of diseases and symptoms, and (2) human diagnostic behavior, revealing critical gaps between medically possible and clinically considered diagnoses. By strategically perturbing latent patient representations, it generates contrastive counterfactuals that highlight rare but plausible conditions—conditions typically overlooked due to cognitive bias. Unlike traditional decision - support tools, our system proactively suggests rare diseases not because they are statistically probable, but because they are cognitively neglected. Evaluated on three rare disease datasets, our approach outperforms standard machine learning classifiers in detecting rare conditions while maintaining strong performance on common diagnoses. Beyond boosting accuracy, it fosters hypothesis - driven reasoning, enhancing both diagnostic precision and clinician learning.