BRAIx: Transforming Breast Cancer Screening with AI

BRAIx: Transforming Breast Cancer Screening with Artificial Intelligence (AI)

Breast cancer is the most common cancer affecting women worldwide. Early detection through regular mammograms saves lives, and programs like BreastScreen Victoria have helped reduce breast cancer deaths significantly. But there's always room to improve how we detect cancer, especially using new technologies like artificial intelligence (AI). 

This project explored how AI could be used alongside human radiologists to improve breast cancer screening. The team tested different ways AI could be integrated into the current screening process, which usually involves two radiologists reviewing each mammogram, with a third radiologist review in if they disagree. 

The research found: 

  • AI can help detect cancer more accurately when used as a second reader or when allowed to make decisions on high-confidence cases. 
  • These approaches improved cancer detection rates by up to 2.5% and reduce unnecessary follow-up appointments (false positives). 
  • Using AI could mean that fewer mammograms need to be read by humans, saving time and resources. 
  • The study also found that automation bias (where humans overly trust AI decisions) can sometimes reduce accuracy, especially in multi-reader settings. 

Outcome

AI has great potential to enhance breast cancer screening, particularly when integrated thoughtfully with human expertise. The BRAIx project has shown that AI can transform the future of breast cancer screening to improve screening outcomes, lower harms and reduce treatment costs.  

Findings from this study have informed implementation of a BRAIx Randomised Control Trial in partnership with SVI, BreastScreen Victoria and BreastScreen South Australia.  

Publications

  1. ADMANI: Annotated Digital Mammograms and Associated Non-Image Datasets 
  2. Comparison of AI-integrated pathways with human-AI interaction in population mammographic screening for breast cancer 
  3. AI-based BRAIx risk score for the intermediate-term prediction of breast cancer: a population cohort study 
  4. Australian healthcare workers' views on artificial intelligence in BreastScreen: Results of a mixed method survey study 
  5. Trust in AI Is a "Fluid Process": Building Trust of AI Through Clinicians' Needs in the BreastScreen Victoria Program-A Qualitative Study 
  6. Developing a Typology of Women's Attitudes Towards AI Use in the BreastScreen Programme-A Qualitative Study With BreastScreen Victoria Clients 
  7. Cross- and Intra-Image Prototypical Learning for Multi-Label Disease Diagnosis and Interpretation 
  8. Progressive Mining and Dynamic Distillation of Hierarchical Prototypes for Disease Classification and Localisation 
  9. Mixture of Gaussian-Distributed Prototypes With Generative Modelling for Interpretable and Trustworthy Image Recognition 
  10. An Interpretable and Accurate Deep-Learning Diagnosis Framework Modeled With Fully and Semi-Supervised Reciprocal Learning 
  11. Learning Support and Trivial Prototypes for Interpretable Image Classification 
  12. Knowledge Distillation to Ensemble Global and Interpretable Prototype-Based Mammogram Classification Models 
  13. BRAIxDet: Learning to detect malignant breast lesion with incomplete annotations 
  14. Multi-view Local Co-occurrence and Global Consistency Learning Improve Mammogram Classification Generalisation 

Lead researchers

Associate Professor Helen Frazer
Clinical Director, BreastScreen St Vincents 
Hospital
 

Partners

St Vincent’s Research Institute

 St Vincent's Hospital

 The University of Melbourne 

The University of Adelaide

Funding

The “Transforming Breast Cancer Screening with AI” (BRAIx) research project was funded by an Australian Commonwealth grant awarded through the Medical Research Future Fund (MRFF) under the Clinical Trials Activity Initiative – 2021 Clinical Trials Activity Grant Opportunity, Stream 5 (Grant No. MRF2023336). 

Quick exit