AI for Health and Care

AI is expected to play a role in addressing urgent healthcare challenges the world is facing, such as shortage of healthcare workers, increasing health disparities, emerging threats from global pandemics, the management of non-communicable diseases (e.g. cancer) and age-related diseases (e.g. Alzheimer’s disease) and the long trajectory for development of new drugs. In Delft, we believe that Artificial Intelligence can enable innovative approaches to health, well-being, and healthcare, and shape the future of medical professionals, patients, and their families.

Artificial intelligence can outperform humans in pattern recognition and machine learning tasks, but they fall short on tasks that require general world knowledge, common sense reasoning, collaboration, adaptivity, responsibility, transparency, and explainability. Our vision is to espouse a "human-centered" model for Artificial Intelligence integrated in a Health & Care ecosystem that is effective, efficient, and equitable. In this model, the integration of human and artificial Intelligence (Hybrid Intelligence) is a core design principle for methods and data (infrastructure).  We will design and develop hybrid intelligence solutions for healthcare, where humans and AI intelligence are optimally combined for AI technology development (e.g., AI model training), control (e.g., inspection, and evolution of AI behaviour), and operation (e.g., humans interact with intelligent agents on common tasks as trusted as a co-worker and companion).

AI Research objectives 

Four AI research objectives have been identified has part of the framework for AI for Health & Care as research focus theme:

AI & Individuals’ Data
Precision medicine exploiting personal data such as omics, medical imaging
Future Health & Care approaches will need reliable, scalable, and affordable data. To obtain this data efforts are needed in generation, enrichment, annotation, integration, and interpretation of multimodal and heterogeneous medical data (think of genomics, metabolomics, transcriptomics, epigenetics, biometrical data, psychological profiles, diagnostic data like MRI scans, etc.), drug-related data (e.g., potential drug-drug interactions, allergies, personal genome), and diagnostic and treatment data.
AI & Population Data
Organize how health is distributed
Population health is about the orchestration of health care, which covers both clinical (intramural), as well as extramural health care, which includes preventive and curative aspects. Key concerns are public health, quality of care, teamwork, inter-professional communication, culture, medical protocols and shared medical ontologies and bodies of knowledge. How to keep health care accessible to all in a greying population? How to organize health care around (possibly conflicting) values? What can we do to orient health care on prevention?
Human-AI Interaction & AI-mediated People Interaction
Behaviours change program, personalized care, consultation room of the future
People interaction is about the meaningful dialog between patients, therapists, doctors, measurements, data, and (semi-automatic) interpretations and summarization to reduce registration burden and improve structured data. In terms of AI challenges, it involves intelligent interaction between AI and individuals, or groups of people. Challenges also include the personalised sensor measurements , and their visualisation and explanation to patients.
AI & Personal medical data sharing and processing infrastructure
Privacy, scalability, etc.
The retrieval, storage, distribution, sharing, and processing of personal medical data pose several infrastructural challenges that can be addressed through novel AI techniques. The storage of sensitive medical data (such as sequenced DNAs) needs to be secured through novel encryption means (such as quantum encryption) to prevent external parties from exploiting this data. The data exchange infrastructure must be in compliance with the data-protection and privacy regulations, while also promoting consensual medical data sharing to promote and facilitate medical research. Tele-medicine (also through Virtual/Augmented reality applications) requires effective and reliable spatial computing and fast communication infrastructures. How can existing cloud-based infrastructure be appropriated or redesigned to facilitate the sharing and retrieval of medical data in rapid ways, e.g. through Edge-AI computing? How to enable the exploration of datasets featuring 100s of features, possibly coming from a single examination?