Background
Cancer is a leading cause of death globally with increasing incidence, breast cancer being the most common cancer in women. Breast cancer is a heterogeneous disease, which is reflected in high variability in outcomes even within well-established subtypes of the disease.
Different aspects of this complexity can be glimpsed from different data modalities including mammograms (radiology) and histopathology slides, molecular information from RNA sequencing, and clinical data from health registries.
In clinical practice, these data are acquired at different stages of the patient’s journey and often manually analyzed and considered in isolation when making clinical decisions. It is beyond human capabilities to integrate the full extent of information and arrive at a detailed and comprehensive view of the entire patient journey.
This project will develop both methodology and AI-models for individual data types (radiology, pathology, molecular, clinical), and models that fuse multiple data types and consider the longitudinal dimension of the patient journey.
The goal is to advance precision diagnostics and decision support for breast cancer treatment.
Research Question
The project focuses on precision diagnostics solutions that offer either prognostic or treatment response predictions.
In addition, the project focuses on rational clinical decision-making that integrates a multitude of parameters, including both routine clinical information and AI-based predictions to provide certified conclusions using causal inference and conformal prediction, and provide an intuitive interface to support decision-making in multidisciplinary teams.
Aim
The overarching aim is to make strategic scientific advances in data-driven multimodal methods to enable true precision diagnostics throughout the breast cancer pathway. The main aims are:
- Development and validation of single- and multi-modal AI-based precision diagnostic models to reduce costs and increase equality in access to advanced diagnostics.
- Focused methodology development in AI-based computer vision, multi-modal AI-based modelling, and data-driven clinical decision-making under uncertainty.
- Development and implementation of data models and infrastructure for multi-modal AI-driven diagnostic research, facilitating transition between research and healthcare.
- Establish the world’s largest multi-site and multi-modal (radiology and pathology images, RNAseq molecular profiling, clinical data) breast cancer study (N >10,000).
Synergy and Team
The project, AID4BC, consists of partners at four sites, with at least two investigators at each site. This is likely the only constellation globally having access to large (>10,000 patients) matched multimodal data across radiology, pathology and molecular profiling and clinical data.
The project has the capacity to perform fully independent validation studies, and to implement prospective validation studies. AID4BC partners have previous experience of developing regulatory-approved medical devices for clinical use, opening clear routes toward clinical translation.
Karolinska Institutet: Provides expertise in predictive medicine and AI-based precision diagnostics, computational and clinical pathology, and cancer epidemiology. KI also contributes large population-representative cohort studies.
Researchers: Main PI, Senior lecturer and Docent Mattias Rantalainen, Prof. Johan Hartman, Dr. Bojing Liu
Lund University: Provides unique data, along with expertise in generating and analyzing molecular data, AI for radiology, pathology and precision-medicine.
Researchers: Prof. Sophia Zackrisson, Associate Prof. Predrag Bakic, Associate Prof. Magnus Dustler.
Uppsala University: Provides expertise in statistical machine methods focusing on causal foundations for internal and externally valid decision-making, as well as efficient learning of multi-modal AI models for large-scale data.
Researchers: Associate Prof. Dave Zachariah, Associate senior lecturer/ Assistant Prof. Jens Sjölund
Linköping University: Offers strong expertise in tackling healthcare adoption challenges for data-driven precision diagnostics. The team brings deep understanding in machine learning and AI-related visualization and access to infrastructure at CMIV.
Researchers: Adj. Prof. Claes Lundström, Associate Prof. Daniel Jönsson
Collaboration with industry and other organizations
Companies Sectra and Stratipath are integral parts of the project. Lund University’s industrial partners include Siemens, ScreenPoint (Transpara), and Collective Mind Radiology. Through the national Analytical Image Diagnostic Arena (AIDA) the project gains a strong connection to Swedish industry and healthcare, as well as with SciLifeLab. Through the innovation environment Swedish AI Precision Pathology (SwAIPP), strong collaboration is established with healthcare, patient organizations and industry.
Impact
AID4BC will have a major impact on several aspects of both the scientific and clinical fronts of breast cancer but will also push the frontiers of AI-based precision diagnostics and causal machine learning methodologies.
In addition to an increase in the understanding of breast cancer and novel biomarkers related to outcome and therapy response, it will drive future breast cancer research and contribute towards transforming clinical diagnosis and management of the disease.
AI-based analyses and decision support hold the promise to reduce costs substantially for precision diagnostics and increase equality in access, ultimately contributing to better outcome for all patients.
The project anticipates multiple high impact publications, with outputs from the project forming the basis for future studies, trials, and translational activities with industry partners.
Cover photo: Photo by National Cancer Institute on Unsplash
