BreaCS is an innovative research project within the healthcare sector that focuses on developing an artificial intelligence (AI) -based clinical decision support system (CDSS) to improve breast cancer treatment selection.
The BreaCS Consortium is a collaborative research effort between four Belgian EUSOMA-certified hospitals (AZ Delta, OLVZ Aalst, AZ Groeninge and GZA) and Endare as external partner.
Objective
BreaCS aims at developing a quasi-realtime A.I.-based clinical decision support system (CDSS) for supporting breast cancer treatment selection. It will combine pathology, radiology and clinical data in uniquely smart A.I. models. This project will examine the potential of a CDSS for breast cancer treatment and at once develop solutions for the known technical issues, as well as multi-center performance evaluation.
Methodology
First concrete objective is harmonizing the data from the 3 participating hospitals (AZ Groeninge, GZA and AZ Delta) into a private virtual cloud which will be setup in Google Cloud, with care for security, privacy and ethical considerations. Only the 3 hospitals will have access to the full datasets.
Second objective is developing three A.I. models:
- Model 1 deals with radiology breast scans (mammogram and/or MRI) and aims at predicting the pathological tumor sizefrom these scans, preoperatively. Human estimation can be up to 3mm off. The aim is to have ± 4 pixels accuracy. It is an important parameter to decide the type of surgery: breast conserving or not?
- Model 2 will preoperatively predict the need for axillary lymph node removal. Given a false positive rate between 10 and 30% according literature, the aim is to be well within 10%. Unnecessary lymph node removal comes with serious side effects for the patients which potentially can be prevented.
- Model 3 aims at predicting the response to neo-adjuvant chemotherapy. Complete pathological response is less than 60%, meaning at least 40% of the patients do not respond and therefore get the therapy unnecessarily. The aim is to achieve precision and specificity > 0.85 to make the model clinically relevant.
Impact and future directions
- By developing these three AI models, the BreaCS project aims to revolutionize breast cancer treatment selection, making it more accurate and personalized. When successful, the outcome of this project is a set of A.I. models which are ready for evaluation in a randomized clinical trial (RCT). The RCT is planned in a succeeding project.
- In Belgium 1 out of 7 women will be diagnosed with breast cancer during her lifetime. If the CDSS proofs to be predictive for the three predefined objectives and models, and given the incidence rate of breast cancer, this project potentially can have a massive impact on medicine and public healthcare. The concepts that investigated in this project on breast cancer can be used in other oncological settings as well. Optimizing treatment plans will result in better outcomes for the patients most likely even at a lower cost.
General info & contact
Keywords (#): Breast cancer, AI models, Multimodal, Multi-institutional
RADar project research lead: Prof. Dr. Peter De Jaeger
RADar project researcher: Dr. ir. Sandra Steyaert
Principal investigator: MD Barbara Bussels
Timeline:
- Project start: 01/10/2022
- Projected project end: 01/12/2042
Status:
Data from all institutions has been collected in a secure cloud infrastructure. Both Endare and RADar are now actively preprocessing and analyzing the data to build AI models.
Publications / presentations:
Abstract submitted and accepted for poster presentation for EBCC14 conference (https://event.eortc.org/ebcc14/)
“Multi-institutional predictive model for axillary nodal involvement using the EUSOMA database”
Partners:
The BreaCS Consortium is a collaborative research effort between four Belgian EUSOMA-certified hospitals (AZ Delta, OLVZ Aalst, AZ Groeninge and GZA) and Endare as external partner.
Funding: Vlaio and the European Union’s NextGenerationEU initiative
RADar - Learning & Innovation Center 