HistoLens: An Interactive XAI Toolkit for Verifying and Mitigating Flaws in Vision-Language Models for Histopathology

Explainability is essential for clinical adoption: HistoLens demonstrates that doctors require more than accurate AI predictions—they need visual explanations (heatmaps showing which cells influenced diagnoses) to understand and trust AI reasoning in high-stakes medical decisions, similar to consulting a knowledgeable colleague.

Interactive questioning improves usability: The system allows pathologists to query AI findings in plain English rather than navigating complex technical interfaces, making AI tools more accessible to medical professionals and keeping the doctor as the decision-maker rather than a passive observer.

AI systems can be trained to focus on relevant information: HistoLens was designed to ignore background noise and distracting elements in tissue slides, demonstrating that developers can actively mitigate flaws by teaching AI models to concentrate only on clinically relevant features—a practical approach to reducing unreliable outputs.

Verification workflows maintain human expertise: By positioning AI as a verification tool rather than a replacement for pathologist judgment, HistoLens preserves the critical role of human expertise while potentially enabling faster, more confident diagnoses through human-AI collaboration.

Transparency enables flaw detection: The toolkit's ability to provide visual proof of AI reasoning allows pathologists to identify when the system makes mistakes or focuses on wrong regions, creating an opportunity to catch and correct AI errors before they affect patient care.