Interactive ai helps radiologists improve image diagnosis accuracy
Interactive image diagnosis using test-time backpropagation from the auxiliary task
Computer Vision and Pattern Recognition
Summary
Medical AI systems try to explain their decisions by showing extra information about findings in images, but sometimes doctors don’t trust these explanations. The authors introduced a way for doctors to give feedback directly to the AI system during diagnosis, effectively asking the AI for improved suggestions based on that feedback. This method uses the AI’s training without needing to train a new system and was tested on breast and lung images, showing it helps make better diagnoses. It also works better than previous approaches that only looked forward in the AI, not adjusting based on feedback.
What this means in practice
- •For radiology teams: Allow radiologists to interact with AI diagnosis tools by giving feedback on auxiliary findings to improve diagnostic accuracy in clinical imaging.
- •For medical imaging software developers: Integrate interactive backward feedback methods into AI diagnosis tools to enhance trustworthiness and accuracy without retraining models.
Authors
Takuro Shimaya
Abstract
Multi-task deep learning has been developed to ensure explainability in computer-aided diagnosis (CAD) systems. These systems simultaneously predict interpretable characteristics of findings in auxiliary tasks to help radiologists assess the reliability of the diagnostic prediction. However, the CAD's opinion would not be referred to when the auxiliary task prediction deviates from the radiologist's expectation. To further leverage CAD's knowledge in the above situation, we proposed a multi-task gradient assist (MGA) method that brings interactivity to multi-task models without additional training. MGA allows radiologists to give feedback to the CAD and ask for additional diagnostic suggestions. Using two datasets of mammography and CT, we show that MGA successfully improves diagnostic accuracy upon feedback on various auxiliary tasks. Compared to an existing method using forward propagation only, MGA performs more robustly and can be combined to further improve accuracy. Our method paves a promising way for integrating artificial intelligence into image diagnosis.