Visualizing Future Breast Cancer Prognosis - Integrating AI Predictions into Screening Programs

Abstract

Sweden has a long tradition of mammography breast screening programs. AI’s integration into healthcare is rapidly expanding. Those AI solutions are mainly aimed at replacing one of the two radiologists who are required to read a mammogram. However, when using the AI solutions available, current breast screening programs have limited ability to prevent fast-growing tumors from becoming incurable. This is especially important in the significant portion of breast cancer cases detected between screenings, called interval cancer. The rate of interval cancer is higher among young women, women with a family history of cancer, as well as black women. In order to address this gap, the TechMed AI startup Aileen Health is developing an AI system to predict interval cancer occurrences, aiming to shift healthcare focus from diagnosing breast cancer to preventing tumors from becoming incurable. This thesis examines radiologists’ interaction with the AI system in order to design an AI solution that builds trust and motivation. More specifically, the thesis aims to answer the research question:

What are key factors for trusting and integrating AI-based predictions into the current mammography screening workflow to support radiologists?

The thesis employs a Research through Design approach to establish requirements and guidelines for designing an AI system applicable to disease progression in medical imaging. It also proposes a visual representation of the AI-generated content through an iterative, collaborative design process. The evaluation identified key factors for establishing trust in AI systems, resulting in 14 requirements and guidelines across four categories: usability, trust, motivation, and technical. The design proposal is now a part of Aileen Health’s product demo and landing page and will be used in further product development. The thesis emphasizes that trust in AI systems is influenced by factors beyond design, such as pre-existing attitudes toward AI, and should be considered together with external factors for comprehensive trust establishment. Overall, the thesis provides insights for implementing AI in healthcare systems, focusing on the design of AI-generated content. By doing so, there is a potential for using AI for predicting disease progression, in cancer treatment and beyond.