Johan Hartman

Cancer remains a leading global cause of death, with rising incidence placing increasing strain on healthcare systems. Breast cancer, the most common cancer in women, is heterogeneous with variability in patient outcomes. To optimize breast cancer care, accessible and cost-effective precision diagnostic solutions are needed to provide clinicians with reliable decision support for treatment allocation. AI-driven approaches for treatment response prediction, prognostic modeling, and phenotyping can enhance patient stratification by identifying subgroups likely to benefit from specific therapies or requiring intensified treatment.Current precision diagnostics relies primarily on molecular diagnostics, which remain expensive and inaccessible to most cancer patients worldwide. AI-based diagnostic solutions for analysis of readily available data modalities offer a potential route towards sustainable precision medicine.This project extends our ongoing research in AI-driven precision pathology, leveraging large studies and deep learning to analyse routine H&E stained histopathology images that provide a rich and underutilized data source for cost-effective precision diagnostics with low implementation barriers. This research aims to advance deep learning methodology for histopathology, develop and validate breast cancer precision diagnostic models, and contribute to increased access to precision diagnostics and improved patient outcomes.

Breast cancer (BC) remains a global health challenge, with 7.8 million new cases annually. Despite treatment and screening advances that have gradually improved outcomes since the 1970s, many patients still do not survive their dissease, creating an urgent need for precision diagnostics to identify high-risk individuals and predict therapeutic responses. The Consortium for AI in Registry-Based Image Epidemiology Research in Breast Cancer (CARE-B) addresses this by combining registry data, histopathology images, and AI to improve BC characterization.CARE-B will establish an internationally unique large (up to 30,000 patients), multimodal multi-site database integrating clinical data, whole slide images (WSIs), and molecular profiles from breast cancer cohorts in Sweden, Denmark, and Scotland. The consortium will develop scalable AI models for cost-effective precision diagnostics, focusing on deep phenotyping of BC subtypes based on routine H&E stained histopathology slides, characterise intra-tumor heterogeneity (ITH), predicting treatment responses and for prognostic stratification.CARE-B will foster collaboration, support junior researchers, and advance epidemiological research and clinical translation. By leveraging AI and registry data, CARE-B aims to significantly impact BC research and clinical diagnostics, creating a foundation for future advances in precision medicine, while also building a research environment that take health-registry research to the next level.

Most breast cancer patients are diagnosed with hormone-dependent (estrogen receptor-positive, ER+) breast cancer and endocrine treatment is standard of care. A unique feature of ER+ disease is that the risk to develop distant metastasis remains stable beyond 5-10 years after diagnosis, and half or more of all metastases will be diagnosed after this initial follow-up. The tumor biological factors underlying long-term risk are poorly understood, and it will remain a considerable clinical challenge in the foreseeable future. We will investigate the influence of standard clinical markers, their intra-tumor heterogeneity, and the heterogeneous ER+ tumor microenvironment, to identify tumor characteristics influencing long-term risk and benefit from endocrine treatment. Novel deep-learning methods will be used and in depth spatial analysis will enable understanding of tumor biology down to single-cell level. We will use unique and large clinical trials with patients randomized to endocrine treatment versus not with complete long-term follow-up. The distinction of long-term risk is essential, since accurate risk prediction allows for individualized treatment, decreases anxiety, and supports aggressive treatment for patients at high risk of fatal disease. Our study has the potential to answer vital questions about the influence of the tumor microenvironment and intra-tumor heterogeneity for long-term risk in ER+ breast cancer, helped by the interdisciplinary expertise in our team.