AI can only Transform Healthcare if Data is Clean, Relevant and Accessible
By Sara Dalmasso, VP Digital EMEA, GE Healthcare
Artificial intelligence (AI) is often positioned as the savior of pressured healthcare systems. Done right, it has the potential to be the technology of the decade, perhaps the century, in healthcare as it can help join up and analyze levels of data that no human brain could process, to recommend, for example, the best course of treatment for an individual patient.
"Creating an interoperable bridge between how we acquire and organize data coming from multiple sources, and the development of AI, has potential to yield a new dimension for modern healthcare"
Ideally, all medical data would be pulled into a system that could recognize patterns based on data from thousands or millions of anonymized patients in similar situations. If AI can be integrated seamlessly into a medical professional’s daily work, it could support decision-making and take on routine tasks, freeing up more time for complex cases.
But before we get to that ideal state, we first have to solve one of the biggest issues: data. As much as 30 percent of data stored the world over is generated in the healthcare industry but it exists in different systems and silos and is not consistent in form. The success of AI is completely dependent on the size and quality of annotated data sets and that the right data is available at the right time.
Groups doing clinical research or developing algorithms, for example, need access to imaging data sets for training and validation. They are looking for specific real-life cases that have been reported on by experts in their field to be able to ‘teach’ a computer algorithm to recognize the difference between healthy and diseased tissue or organs.
Let’s take the example of lung nodule detection and classification algorithms. Rather than overwhelming developers with millions of random chest x-rays or scans, we should provide them with more specific data sets, such as, “non-smoker males diagnosed with lung cancer below the age of 40 years”. Extraction of the right data will accelerate the potential of AI.
Access To Clean and Diverse Data Helps Re- Imagine Medicine
We should consider having reports stored and demographics parsed into specific database fields so that the data can be queried and segmented in any and every way. This means that searches and extraction of data can be run on patient population queries or on diagnosis type. Standards-based data can then be connected through advanced analytics to identify and eliminate inefficiencies and to improve performance. This would help clinicians in hospitals now and support future innovations.
A Vendor Neutral Archive (VNA) can help. It houses medical images and files of clinical relevance from across the healthcare enterprise—drawing data from disparate systems, across multiple specialties using international standards such as IHE cross-enterprise document sharing. Accessed via a single standard interface, it can unify the clinical-‘ologies’ for a complete picture of patient data.
The efficient exchange of patient data information isn’t about an IT department buying as many solutions from the same vendor as they can, it is about various systems communicating smoothly with each other to avoid missed opportunities in the patient care continuum. To be ready for the advantages of emerging health technologies such as AI, healthcare also needs solid interoperability foundations in place for future developments. VNA can help in this aspect by enabling better collaboration between all stakeholders in patient care.
Creating an interoperable bridge between how we acquire and organize data coming from multiple sources, and the development of AI, will yield a new dimension for modern healthcare. The amount of data we see today is only a fraction of what will exist in five years. By managing patient data better and using analytics, we can gain more control of outcomes.
One way we’re already starting to glimpse this is as we co-develop AI applications with leading academic institutions, hospitals and health systems, focused on key care areas like neurology and stroke, and workflow efficiencies such as high x-ray reject rates. One early example of an algorithm under development is a solution for pneumothorax, or a collapsed lung. The algorithm will be focused on teaching machines to distinguish between normal and abnormal scans so clinicians can prioritize and more quickly treat patients with pneumothorax, which can be a life-threatening condition.
With the EMRAD (East Midlands Radiology Consortium) in the UK, we have created a radiology imaging network with shared access to more than 3 billion images in one VNA for the benefit of almost 5 million patients and clinicians. As EMRAD explores how AI can support breast screening, having data easily accessible in the VNA underpins this work to validate the algorithms’ effectiveness. We are working with a national consortium of NHS Hospitals, charities and SMEs convened by Oxford University, which will undertake research to develop and implement new AI based solutions to help speed up and improve accuracy in detection of conditions including lung cancers, cardiovascular and liver disease. Our Centricity Clinical archive is a key part of the data collection and storage strategy and will connect the 15 NHS trusts that are part of the project.
As the amount of healthcare data held by hospitals continues to increase, interoperable data storage and VNA systems will become ever more important.