This thesis is inspired by the large corpus of AI research done focused on screening for skin cancer. The importance of such screening is underlined by the rising numbers of skin cancers diagnosed each year paired with the lack of health care and unawareness of people about skin cancer (e.g., Urgurlu, 2026). Thus, application such as SkinVision using AI to help people in their self-examinations become more and more widespread (Brancaccio, 2024). While their accuracy currently is already high, rare or untypical presentations of skin cancer, that are underrepresented in the training datasets, can still lead to misdiagnosis (Frasier, 2025). For example, darker skin tones are underrepresented. For such aspects studies already explore solutions such as generating images for these skin tones to increase the dataset (Abhari, 2023; Rezk, 2022). 
While such technical approaches are promising, we want to take a different approach focusing on the human. Current studies in the field of human-AI interaction focus more on supporting the users reasoning process instead of on providing recommendation, thus, the aim is to augment the users skills during interaction (e.g., evaluative AI (Miller, 2023); forward reasoning (Reicherts, 2025)). This may be especially of interest, as it could educate the user about typical warning signs (ABC rules) and raises awareness of the topic. Further, if untypical cases the AI may be less equipped to handle occur, user may have enough knowledge and thus confidence to not be swayed.

Summarized, the focus of this work is on skin cancer self-checks. The goal is to explore how publicly available data (HAM1000, ISIC) can be used not only to train an AI model, but to specifically help in assessing the ABC warning signs. Some research already designs the explanation for the AI prediction based on the ABC rules. However, how such can be used by humans and whether how they are calculated matches how humans understand the rules remains to be investigated. This is where your work starts: Explore the possility space of what can be done with the data to provide information on the ABC rules, conduct interviews to capture how people utilize the ABC rules, and then conclude how to technically match the user expectation. 

Procedure

To reach this goal, the following steps can be taken:

1. Get to know the literature. What are current developments regarding skin cancer, AI and the ABC rules? What are technical approaches? Were user studies conducted?
2. Conduct interviews with at least around 5 individuals and analyse them using a content analysis with MAXQDA,
3. Derive their needs and use them to decide what technical approach is the most promising in supporting them,
4. Create a functional prototype (at least with basic AI back-end and possible ABC rule explanations - which do not need to be technical but compliment basic AI outputs) 
5. Conduct interviews with at least 5 individuals to evaluate the updated prototype.

References (available on campus/via VPN)

• Ugurlu, Z., Işık, S. A., Balanuye, B., Budak, E., Elbaş, N. Ö., & Kav, S. (2016). Awareness of skin cancer, prevention, and early detection among Turkish university students. Asia-Pacific journal of oncology nursing, 3(1), 93-97. https://pmc.ncbi.nlm.nih.gov/articles/PMC5123539/
• Frasier, K., Hash, M. G., Werpachowski, N., & Fritts, H. (2025). The blind spots of artificial intelligence in skin cancer diagnosis. Dermis, 5(2), 1-7. https://www.jdermis.com/full-text/the-blind-spots-of-artificial-intelligence-in-skin-cancer-diagnosis 
• Abhari, Julian  and Ashok, Ashwin . 2023. Mitigating Racial Biases for Machine Learning Based Skin Cancer Detection. In Proceedings of the Twenty-fourth International Symposium on Theory, Algorithmic Foundations, and Protocol Design for Mobile Networks and Mobile Computing (MobiHoc '23). Association for Computing Machinery, New York, NY, USA, 556–561. https://doi.org/10.1145/3565287.3617639
• Rezk, E., Eltorki, M., & El-Dakhakhni, W. (2022). Leveraging artificial intelligence to improve the diversity of dermatological skin color pathology: protocol for an algorithm development and validation study. JMIR Research Protocols, 11(3), e34896.
• Miller, T. (2023, June). Explainable ai is dead, long live explainable ai! hypothesis-driven decision support using evaluative ai. In Proceedings of the 2023 ACM conference on fairness, accountability, and transparency (pp. 333-342). https://dl.acm.org/doi/pdf/10.1145/3593013.3594001
• Reicherts, L., Zhang, Z. T., Von Oswald, E., Liu, Y., Rogers, Y., & Hassib, M. (2025, April). AI, help me think—but for myself: Assisting people in complex decision-making by providing different kinds of cognitive support. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (pp. 1-19).
• Brancaccio, 2024, Artificial Intelligence in Skin Cancer Diagnosis: A Reality Checkhttps://linkinghub.elsevier.com/retrieve/pii/S0022202X23029640