Artificial intelligence has the potential to change healthcare at almost every level. But it’s not meant to replace clinicians — it’s about bringing together and integrating further healthcare expertise with advanced technologies to discover medical breakthroughs, which will ultimately improve patient care. A practical way to look at it is that AI can serve as a data-driven research or clinical assistant, both reflecting the experience of a collective clinical evidence.
The vast amount of data generated and collected by stakeholders in healthcare comes in many different forms, including insurance claims, physician notes, medical records, medical images and pharmaceutical research and development — not to mention conversations about health in social media and information from wearables and other monitoring devices.
Since this vast amount of information is meant to deliver care, it is digitally structured from a patient-centric perspective. There lies a formidable challenge in making real-world data (RWD) more accessible for insight generation related to personalized outcomes such as therapy response, surgery recovery, and disease progression. Although it may seem counterintuitive at first, there is a need to have a drug-centric understanding of response if one’s objective is to personalize treatments for a given patient. This requires restructuring all real-world patient-centric data by placing the drug at the centre.
Rethinking accessibility in health care
Hospitals and healthcare organizations (HCOs) are increasingly relying on digital technologies, so the ability to access, exchange and leverage data becomes critical to patient care. Accordingly, HCOs have embarked on digital transformation strategies, mostly centered on data maturity models. Yet these models have been established before the advent of deep learning, and do not support discovery processes. Data maturity models do not directly address the issue of knowledge accessibility, particularly in the context of AI learning from RWD. That is about accessing knowledge locally and then furthering that collectively. Accordingly, there is a need to increase the flow of information across healthcare siloes from both the human and data perspective. So, although data maturity strategies focus on data access and centralization, it falls short of unlocking knowledge and then furthering that knowledge collectively.
The Imagia EVIDENS™ AI platform is addressing the shortcomings of such data maturity models by indexing RWD(such as labs, reports and images), extracting RWE (such as outcomes of interest), and structuring information to allow the design of AI-first and evidence-based clinical research. Ultimately, Imagia’s EVIDENS™ allows clinical researchers to perform rapid hypothesis validation from local RWD, and scale promising discoveries through privacy-preserving federated learning.
Prior to federating knowledge, it starts with producing local evidence for a clinical hypothesis, in a given hospital, understanding their specific institutional practices, policies and languages, and structuring their data with the particular characteristics of their equipment for their specific population and sub-populations. This is critical in order to cross the translational gap of testing medical hypotheses and have an effective strategy for clinical adoption, through the translation of well-qualified (robust and reproducible) AI-driven discoveries such as new biomarkers, the development of new therapies and diagnostics.
Enabling clinical adoption by design
The way Imagia thinks about how AI can be used by clinicians in a hospital setting is similar to what we have witnessed in oncology with the concept of tumor boards — where healthcare experts from various oncology-related disciplines come together to define the best care strategy for a given patient, making treatment more personalized and ultimately more effective.
Tumor boards have indeed addressed some significant human and organizational barriers to personalized medicine. Consider the case of lung cancer patients – the care pathways cross the disciplines of radiology, surgery, pathology and oncology, in that traditional order. In order for oncologists to leverage advances in targeted therapy, sufficient tissue samples should be obtained at time of initial biopsy to test for clinically-actionable mutations long before the oncologists get involved in patient care. This organizational limitation contributes to a reported 40% re-biopsy rate, partially explained by the lack of early coordination to the downstream aspects of care (source: MISC – Vol.25 2017 – Human Futures).
Tumor boards have proven to be highly effective as they are all about information accessibility and sharing. With a tumor board, a multidisciplinary team of healthcare experts meet together and see patients at the same time, early in the patient care pathway. This results in the sharing of data, creating an efficient “interface” to grow knowledge based on real-world evidence, and a timelier, more informed response and personalized patient care.
AI, however, isn’t meant to replace any of these experts; rather, AI is another ‘expert’ in the room, providing another ‘collective voice’ to the conversation, and improving teamwork within organizations. And such ‘collective voice’ is the result of a collective transformation of RWD into practical real-world evidence. AI also offers the benefit of ‘repeatability,’ because decisions are made without any bias of sentiment or emotions, which is helpful to clinicians after long, stressful shifts making life-and-death decisions — not to mention the potential to improve the administration-clinician-patient relationship.
At Imagia, we want to do with AI what has been done in oncology with tumor boards, by taking accessibility to another level where clinicians can further share knowledge supported by real-world evidence and collaborate for better decision-making. We want to empower clinicians to improve patient outcomes. To that end, our platform allows clinicians to be the driving force of the design process of AI solutions using RWD.
How it works
Using AI, we create a representation of patient information from all data sources at the hospital — everything from clinical reports to labs, prescriptions and electronic health and medical records. Once data is in this knowledge base, clinicians can quickly ‘rearrange’ the perspective on any of the data.
EVIDENS™ user interface is purpose-built for clinicians to discover groups of patients with similar characteristics — almost like a Google query on healthcare data. We then use natural language processing/understanding (the technology behind how computers interact with natural human speech) with engineering innovations in big data management to constantly manage ever changing unstructured RWD, with the aim of quick scalability as the system identifies and restructures medical facts dynamically.
We then map this growing body of knowledge to clinician-friendly semantics, such as demographics, diagnoses, procedures and medications to ease the query process and establish patient cohorts for research purposes. Clinicians can build patient cohorts associated with diagnostic and treatment responses and link them to achieve AI-friendly data to learn from.
The next steps usually require machine learning specialists to design, select and tune AI models, a process often referred to as “the art of model & hyperparameter selection”. Yet such expertise is limited to a selected few, particularly for healthcare AI. To alleviate AI resource scarcity, Imagia traded access to AI experts for computing resources; effectively, we have designed SELF, Imagia’s Self-Evolving Learning Framework solution that automatically creates novel AI architectures for a given AI-ready data, and ‘trains’ it, similarly to what an AI scientist would do. It’s bias-free, purely data-driven, it’s relatively fast and it can scale. For example, SELF achieved a 90 percent accuracy rate for the complex task of detecting lung cancer in 3-D computed tomography (CT) scans from a public dataset ; the foundational breakthrough here is that the optimal AI model was automatically discovered through our data-driven process, and achieve state-of-art performance compared to expert-designed AI architectures.(https://financialpost.com/pmn/press-releases-pmn/business-wire-news-releases-pmn/imagia-partners-with-top-us-and-canadian-hospitals-to-facilitate-ai-accelerated-healthcare-discoveries)
This allows clinicians to autonomously engage in different discovery processes and validate clinical hypotheses, without requiring a team of AI scientists by their side.
Scaling to a collaborative ecosystem
Everything starts with clinicians, managing patient care pathways in their local environment. By bringing together their expertise with advanced AI, Clinicians are positioned best to unlock the potential of real world data in their organization. Their discoveries will reveal real world evidence that supports and improves on best practices, to achieve collaborative medical breakthroughs, and ultimately deliver on the promise of personalized patient care.
There lies a new paradigm: by shaping reproducible research from real-world data, the platform EVIDENS™ offers the right foundation to scale open innovation, while preserving institutional data ownership, governance, and ensuring patient privacy. Building on that unique innovation infrastructure, we are delivering federated learning capabilities across an international network of like-minded hospitals & AI academic institutions, to unlock unprecedented clinical breakthroughs.
Blog Jun 7, 2021
Genomic data has the potential to be clinically useful, but its use today is very limited – this potential has not been realized. Imag...
Press May 18, 2021
Imagia boosts its Board of Directors with an experienced biotech industry leader with strong dual domain expertise in technology and healthc...
Publications May 16, 2021
Mohammad Havaei, Ximeng Mao, Yiping Wang, Qicheng Lao. Medical Image Analysis, 2021, 102106, ISSN 1361-8415.
Please complete the form in order to direct your request to the appropriate department, and we will reach out as soon as possible.