A Two-Phase Visualization System for Continuous Human-AI Collaboration in Sequelae Analysis and Modeling

Yang Ouyang,Chenyang Zhang,He Wang,Tianle Ma,Chang Jiang,Yuheng Yan,Zuoqin Yan,Xiaojuan Ma,Chuhan Shi,Quan Li

2024-07-20

Abstract:In healthcare, AI techniques are widely used for tasks like risk assessment and anomaly detection. Despite AI's potential as a valuable assistant, its role in complex medical data analysis often oversimplifies human-AI collaboration dynamics. To address this, we collaborated with a local hospital, engaging six physicians and one data scientist in a formative study. From this collaboration, we propose a framework integrating two-phase interactive visualization systems: one for Human-Led, AI-Assisted Retrospective Analysis and another for AI-Mediated, Human-Reviewed Iterative Modeling. This framework aims to enhance understanding and discussion around effective human-AI collaboration in healthcare.

Human-Computer Interaction

What problem does this paper attempt to address?

The paper attempts to address the issue of how to effectively conduct complex medical data analysis through Human-AI Collaboration in the healthcare field, particularly focusing on the analysis and modeling of hormone-related diseases and their sequelae. Specifically, the paper points out the following problems with existing AI models when dealing with complex medical data:

1. **Simplified Human-AI Collaboration Dynamics**: Existing research often views fully developed AI models as auxiliary tools or reminders, ignoring their limitations in complex medical data scenarios.
2. **Data Complexity and Feature Definition**: Complex medical data requires precise feature definition and detailed input labels, which usually necessitates the involvement of medical experts.
3. **Transparency and Interpretability**: AI models lack transparency and interpretability, making it difficult for doctors to trust their recommendations.
4. **Role Transition and Task Allocation**: In different stages of analysis, the roles and task allocation between humans and AI need to be clear to optimize collaboration effectiveness.

To address these issues, the paper proposes a two-phase interactive visualization system framework aimed at enhancing Human-AI Collaboration in the medical field. The specific goals include:

- **Phase I**: Establishing connections between diseases, medical events, and outcomes through retrospective analysis, covering AI-assisted risk identification. This phase is human-led with AI assistance to improve the efficiency and accuracy of the analysis.
- **Phase II**: Integrating AI models into the final analysis process through iterative modeling, optimizing the models based on feedback from medical experts. This phase involves AI building the models, with humans reviewing and providing feedback to ensure model transparency and interpretability.

Through these two phases, the paper aims to improve the quality and efficiency of medical data analysis and promote effective collaboration between medical experts and AI systems.