Skip to main content Skip to navigation

Enhancing Radiological Diagnosis: A Collaborative Approach Integrating AI and Human Expertise for Visual Miss Correction

Project Overview

The document explores the innovative use of the Collaborative Radiology Expert (CoRaX) system, which combines AI technology with human expertise to enhance diagnostic accuracy in chest radiology. By leveraging eye gaze data and analyzing radiology reports, CoRaX effectively identifies and rectifies perceptual errors made by radiologists, leading to improved diagnostic performance. The findings demonstrate that CoRaX not only enhances the accuracy of diagnoses but also serves as a valuable educational tool, particularly for training novice radiologists. This dual application of CoRaX underscores its potential to transform both clinical practice and educational methodologies in the field of radiology, showcasing the significant benefits of integrating generative AI into medical education. Through its implementation, CoRaX exemplifies how generative AI can facilitate better learning outcomes and support the development of critical diagnostic skills in healthcare professionals.

Key Applications

Collaborative Radiology Expert (CoRaX)

Context: Educational use for training inexperienced radiologists and improving diagnostic accuracy in chest radiology.

Implementation: The system utilizes eye gaze data and radiology reports to identify perceptual errors and enhance decision-making processes.

Outcomes: The system corrected 21% of identified perceptual errors and achieved a Total-Usefulness score of 84% in aiding diagnostic performance.

Challenges: The system faces challenges in fully encompassing the spectrum of perceptual errors in genuine radiology reports and slight misalignment between eye gaze movements and report transcription.

Implementation Barriers

Technical Barrier

Inherent errors caused by misalignment between eye gaze movements and report transcription.

Proposed Solutions: Improving integration techniques and refining data collection methods for better alignment.

Data Limitations

The simulated error dataset may not fully represent the wide range of perceptual errors in actual clinical practice.

Proposed Solutions: Expanding training datasets and incorporating real-world error data to enhance system training.

Project Team

Akash Awasthi

Researcher

Ngan Le

Researcher

Zhigang Deng

Researcher

Carol C. Wu

Researcher

Hien Van Nguyen

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Akash Awasthi, Ngan Le, Zhigang Deng, Carol C. Wu, Hien Van Nguyen

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies