Peter Bosman’s project proposal DAEDALUS (Distributed and Automated Evolutionary Deep Architecture Learning with Unprecedented Scalability) has been granted by NWO TTW in its Open Technology Programme.
Especially due to deep learning, Artificial Intelligence (AI) is pervading society because of huge innovation potential. This includes the medical domain as well. Together with research partner Leiden University Medical Center (LUMC) and industry partners Elekta and Ortec Logiqcare, Bosman (project leader) aims to develop new technology to automate the search for the most suitable deep neural network architectures to tackle a specific task in a way that scales much better than existing approaches. This will be done by fusing promising ideas on neural architecture search with key strengths of another form of AI: a relatively recent novel type of evolutionary algorithm known as ‘GOMEA’, which has been developed along a key long-running research line of Bosman on the design of scalable model-based evolutionary algorithms.
While the technology will generally be widely applicable, a key motivation behind the project are applications in the medical domain. Therefore, the researchers specifically also want to develop variants of the new technology for when data is spread across multiple locations. This is important because for state-of-the-art results, much training data is typically needed, while individual hospitals typically only have limited data and sharing patient data between hospitals is often non-trivial. Solutions will be created for the case when data can be pooled in one location as well as when only information on learned models can be shared.
Real-world utilization and demonstration of the potential of this new technology is aimed to be achieved within the project runtime through two applications in radiation oncology: auto-contouring of medical images and predicting dose distributions for brachytherapy cancer treatment planning. Both applications have the potential to further innovate the research that Bosman is also involved in concerning the accurate, insightful, and automatic construction of brachytherapy treatment plans via AI. Software developed along that research line was recently used for the first time in the Amsterdam UMC to treat a prostate cancer patient.
The research team will consist of 1 postdoc, 1 PhD student, and 1 scientific programmer for CWI as well as 1 PhD student based at LUMC. Furthermore, international collaborations are planned with renowned researchers at the University of Texas at Austin and the Virginia Commonwealth University in the USA, the University of Queensland in Australia, and the Victoria University of Wellington in New Zealand.
Information on NWO website
Peter Bosman is senior researcher at CWI's Life Sciences & Health group