From information about a patient’s family history and readings from their bedside monitors to x-ray images, blood test results, and the medication they’re taking, there’s no shortage of data in the modern healthcare setting. Intelligent algorithms are turning this mass of data into easy-to-use, actionable information to help doctors make faster, more accurate decisions, and even harnessing it to support predictive diagnoses and proactive treatment plans that incorporate highly personalized care.
While data has always been at the heart of healthcare, rapid digitalization means that nowadays any interaction with a healthcare professional generates data in one form or another. Increasingly this data comes not only from hospital devices like scanners or bedside monitors, but from devices worn at home too. While gathering all this information on an individual patient in one place presents a number of challenges, not least from the point of view of privacy and security, in principle doing so gives doctors a complete picture of that person’s medical history. But it’s just that, history, and it can’t predict the future. What if we were talking about information from thousands or even hundreds of thousands of patients, complete with genetic information, environmental factors, and data on things like activity level and even alcohol and smoking habits, as well as outcomes from diseases and interventions. That amount of data becomes an incredibly powerful tool for aiding diagnoses, predicting risks related to future health, and tailoring care according to the needs of individual patients. But only if you have good enough data-analysis algorithms to make sense of it all.
Information overload
The problem in the past has been that we have had lots of data available, but it has been siloed in different databases, making it incredibly hard to combine for the purposes of making an objective diagnosis. Everyone is different, and we don’t have the luxury of an exhaustive data bank that records every possible condition, which makes setting reliability thresholds difficult. Depending on the disease, the available data might be valid for an accurate diagnosis in around 80% of patients, but a worried patient at the doctor’s office doesn’t know if he or she is part of that 80% or not, leaving them with a feeling of insecurity. To give patients more certainty and doctors more accurate support, we need different kinds of information, and this is where resources such as biobanks and genetic data can be extremely valuable.
Algorithms to the rescue
Gathered in one place, this incredibly vast and complex amount of variables can seem like an impossible puzzle to crack. But this is where algorithms come to the rescue. The patient outcome can be a disease occurring, for example, and with lots of the same occurrences a pattern might emerge that the healthcare professional can interpret. But this would not be possible without the help of an AI algorithm that is able to extract patterns, variables, and combinations of variables to present the most likely diagnosis or help predict what a patient’s future holds. The more data available, the more complete the picture and the smarter and more accurate the predictions will be.
Directly involving the doctors and nurses themselves is critical when developing an algorithm for healthcare applications. Combining patient data with technologies like machine learning and artificial intelligence can quite easily produce a lot of personalized information, but it only becomes actionable if it is presented to healthcare professionals in a way they can quickly and easily understand. This is why you need to combine the expert skills of data scientists, mathematicians, usability experts, and programmers with input from the healthcare professionals who will use the application in their everyday work.
Understanding the needs of all stakeholders
VTT has been active in the healthcare arena for more than 20 years, working with doctors and hospitals to develop intelligent algorithms that provide vital decision-making support for medical professionals working on neurological disorders like Parkinson’s and dementia as well as cardiac diseases and cancer. Furthermore, we co-operate with partners from the healthcare industry that will integrate these algorithms into new tools and products.
“Analyzing health data presents a very specific set of challenges – both technical and non-technical – and to answer these challenges it is essential to understand the needs of both healthcare professionals and medical device manufacturers alike,” says Professor Mark Van Gils, Artificial Intelligence in Wellness and Healthcare at VTT. “Our approach is always to think from the real-world perspective – can this really be implemented and what practical issues need to be taken into consideration. For example, we might need to trade off algorithm performance against applicability.”
Putting theory into practice
Companies approach VTT with a variety of needs. At the most basic level, a company or hospital may have a massive amount of data but lack the knowhow to process it. With secure access to their databases, VTT can transform the millions of data points into actionable information. Alternatively, a medical device manufacturer might want to cooperate in developing an algorithm to predict a certain disease or for a certain group of people. Taking things a step further, a customer collaboration or research project might result in the kind of viable product that leads to the creation of a spin-off company or joint venture.
One example of how data has been transformed into a viable product is VTT’s collaboration with Odum Oy, a Finnish provider of digital health solutions, on their AlvinOne AI-based application. The app can predict a user’s risk of becoming ill, perform electronic health evaluations, and provide personal feedback on how to proceed. The technology behind it is based on a machine-learning algorithm developed as part of a long-term collaboration with VTT and is a good example of the kind of solution that can have tangible real-world benefits. For example, by providing users with actionable information on their health, AlvinOne can positively affect people’s behavior, increase wellbeing, and improve workplace productivity.
More time for doctors to be doctors
The technology around artificial intelligence and machine learning is evolving all the time. Self-learning neural networks will continue to speed up the processing of ever-larger amounts of data, which will lead to ever-faster and more accurate diagnoses. That’s the big picture. But on a day-to-day basis, smart algorithms provide doctors with improved support for clinical decision-making and help them focus on the things that really matter by automating less complex tasks such as X-ray assessment, enabling them to spend more time with patients.
“The aim of technologies based on smart algorithms is not to replace the doctor or other medical professional, or make decisions on their behalf, but rather to provide them with a better level of support for clinical decisions,” says Van Gils.
