Researchers developed a deep learning system using eye-tracking data to enhance AI-powered biopsy image interpretation.
Key Details
- 1The Pathology Expertise Acquisition Network (PEAN) was trained using expert pathologists’ eye movements while reviewing 5,881 skin lesion slides.
- 2PEAN-C achieved 96.3% accuracy and 0.992 AUC on internal data and 93.0% accuracy and 0.984 AUC on external test data, exceeding the next best AI model by 5.5%.
- 3Using pathologist gaze data, the model identifies relevant tissue regions with much less manual labeling burden than traditional pixel-wise annotation.
- 4Integrating PEAN-generated data with other models improved classification accuracy and AUC, statistically significant by paired t-test (p = 0.0053 and 0.0161).
- 5The aim is to scale to personalized diagnosis and multimodal predictive models, leveraging low-cost data collection from expertise monitoring.
Why It Matters
This study demonstrates a scalable, low-burden way to transfer human diagnostic expertise into AI tools, potentially enhancing accuracy and acceptance in digital pathology and, by extension, inspiring innovations in radiology imaging AI with analogous annotation challenges.
Source
EurekAlert
Related News
•EurekAlert
AI-Powered OCT Enables Rapid 'Optical Biopsy' for Early Endometrial Cancer Detection
A team at Washington University has developed a catheter-based 3D OCT system with AI to quickly and noninvasively detect early endometrial cancers.
•EurekAlert
AI Clinical Reasoning in Diagnostics and Digital Fatigue in Healthcare
Recent JMIR features explore large language models in clinical diagnostics and digital fatigue among healthcare professionals.
•EurekAlert
KAIST, MIT, Microsoft Develop Efficient AI Image Upsampling for Robotics
KAIST, MIT, and Microsoft have created 'Upsample Anything,' a training-free AI method to restore high-resolution visual data from compressed images with up to 16x improved GPU memory efficiency.