Dermatologist-like explainable AI enhances melanoma diagnosis accuracy: eyetracking study

Comprehensive Summary

This study investigates how explainable artificial intelligence (XAI) systems can improve dermatologists’ diagnostic accuracy for melanoma. 76 dermatologists participated in a two-phase reader study where they diagnosed dermoscopic images using AI and XAI support. Eye-tracking devices were used to determine how dermatologists allocate their attention to XAI explanations. The results showed that the diagnostic accuracy increased by 2.8% points when dermatologists used XAI as compared to AI (from 79.9% to 82.7%). The researchers conclude that XAI can enhance diagnostic accuracy without needing to depend on clinician experience. Additionally, dermatologists showed higher visual engagement and cognitive effort when disagreeing with AI diagnoses, showing that there was deeper analysis during challenging cases.

Outcomes and Implications

Being able to identify early and accurate melanoma is crucial for survival. XAI provides a way to strengthen diagnostic reliability. This study demonstrates how XAI systems can help dermatologists make faster and more accurate diagnoses, while reducing errors and uncertainty. However, further validation in the real world needs to be studied before it can be fully implemented in diagnostic systems. The authors also acknowledge the drawbacks of webcam-based eye tracking systems as compared to dedicated ones.