The ASCAPE Framework aims to provide a means for hospitals and other Healthcare Providers across Europe (and the world) to provide the benefits of Artificial Intelligence to the care of cancer patients. Advanced analytics and facilitated monitoring of cancer patients’ Quality of Life, will be showcased in two project pilots on breast cancer and prostate cancer.
ASCAPE bridges the world of Cloud Computing, Big Data and Machine Learning (ML) with the world of healthcare information systems and sensitive patient data. A typical cloud-based architecture provides the capability to feed Machine Learning algorithms with data collected from different sites. However, in the medical domain, data protection regulations and hospital policies restrict the movement and processing of patient data by third parties in remote infrastructures beyond a hospital’s control. On the other hand, a typical local processing-based architecture would feed Machine Learning algorithms only with local data thus severely limiting their potential. ASCAPE aims to avoid the limitations of both pure local-processing and pure cloud-processing approaches, offering ways in which ML models can be trained on data from multiple hospitals without the hospitals revealing their patients’ data to a third party.
To achieve this goal, the ASCAPE Framework will be built on the basis of an edge computing architecture supporting privacy-preserving ML technologies. The medically relevant data of a hospital’s patients will be processed by ASCAPE’s algorithms at each hospital’s ASCAPE Edge Node, on hardware that will be part of the hospital’s ICT infrastructure and under its control. Doctors at the hospital will have access to ASCAPE Edge-provided AI analytics about their patients offered via graphical user interfaces that present the results in a user-friendly, actionable and meaningful manner.
The ASCAPE Edge Node offers the advantages of local processing of patient data. The ASCAPE Cloud offers the ability of AI knowledge to be collaboratively built between hospitals’ Edge Nodes on the basis of two different technologies. Hospitals can choose if and how they contribute to the maintenance of shared ML knowledge (models).
- Federated Machine Learning: The ASCAPE Cloud and participating ASCAPE Edge Nodes collaborate for training models on the basis of the data of all participating ASCAPE Edge Nodes. ML models and not sensitive patient data are exchanged between the ASCAPE Edge Nodes and the ASCAPE Cloud. The application of ML models to patient data (model inference) for the purpose of producing ASCAPE analytics happens on the hospital’s ASCAPE Edge Node.
- Homomorphic Machine Learning: A number of hospital ASCAPE Edge Nodes contribute knowledge in the form of homomorphically encrypted data which the ASCAPE Cloud cannot decipher. However, the ASCAPE Cloud may build models from the homomorphically encrypted data. It can also use these models to perform model inference on encrypted data, producing homomorphically encrypted results, which, again, it cannot decipher. These results are sent to the ASCAPE Edge Node that requested them where they are deciphered (inside the hospital’s ICT infrastructure) and presented to the hospital’s doctors.
ASCAPE will also employ Differential Privacy, Outlier Detection and state-of-the-art security mechanisms on both the Edge and the Cloud and aim to provide a robust foundation for the large-scale collaborative building of AI knowledge (ML models) that will better support doctors and benefit cancer patients in the future. ASCAPE also brings to the table Explainable AI technologies in order to be able to provide doctors with supporting information for any AI predictions and recommendations, helping them better understand and trust the ASCAPE AI. Finally, the ASCAPE Framework aims to support minimal effort integration with existing hospital information systems and does not only place emphasis on simple APIs but also on providing tools that can bring in information from third-party sources, such as open data and wearable devices, always paying attention to privacy issues.
The ASCAPE architecture (Figure 1) follows a micro-services architectural approach for both the ASCAPE Edge Node and for the ASCAPE Cloud parts of the Framework, with an array of micro-services providing functionality for the synchronisation of patient data (Edge), the training of ML models (Edge & Cloud), computing AI results (Edge & Cloud), training explainable surrogate models for the purposes of explainability of AI results (Edge & Cloud).
The ASCAPE Framework architecture covers all areas of functionality necessary to achieve the desired effect of integrating ASCAPE into clinical practice. The Framework has been designed to work with a wide range of information systems, to allow different applications to be supported beyond the breast cancer and prostate cancer applications foreseen in the pilots, to interoperate with different kinds of devices and open data sources, and even to support different ML technologies. It forms a solid foundation for upcoming technical work in the project and an important step towards the realisation of the ASCAPE vision of large-scale collaborative building of medical AI knowledge and its application to improve patient care.
Blog entry prepared by INTRASOFT International SA