What contribution can artificial intelligence make to medicine?
«The world of medicine generates an enormous amount of information. According to one estimate, healthcare data will exceed ten thousand exabytes (more than ten trillion gigabytes) in 2025. Research in the healthcare sector, in turn, continues to grow: PubMed, the main biomedical scientific literature search service, is enriched with one million publications every year. These are orders of magnitude difficult to understand and unimaginable to analyze manually. First of all, artificial intelligence can and will make an essential contribution in managing, interpreting and making new discoveries from this data.
Overall, AI can make healthcare processes more efficient, leading to earlier diagnoses and greater personalization of treatments. A doctor who has gained experience seeing thousands of patients has a greater diagnostic ability than a novice colleague. Similarly, by processing millions of data (texts, x-rays, ECGs, electroencephalograms, etc.) we could identify characteristics or interpret signals that we do not grasp today. The analysis of this mass of information allows us to recognize patterns, unique characteristics or signals that may escape the individual.
AI can foster precision medicine and support doctors in decision-making. Another major application, of course, is the development of new drugs. In short, artificial intelligence will prove invaluable at all levels. At the SUPSI Institute of Digital Technologies for Personalized Healthcare (MeDiTech), we work primarily at the human level.
Will we be visited by “Doctor AI” in the future? Can we trust her diagnoses?
«The use of AI in the diagnostic field has great potential, but it will be essential to correctly direct its development. Many have asked ChatGPT for medical advice, but the result can be dangerously false information. It is undesirable to follow this direction.
A known problem, or current limitation, of the most popular tools is that they don't tell us the source of their information; they can be victim to hallucinations and bias . A well-documented case involves a hospital group in the United States that relied on artificial intelligence to choose less healthy patients for stronger treatment. The problem is that the algorithm had learned that treatment needs were derived from the costs generated by patients: the more you spend, the sicker you are. In the US, African Americans don't have much trust in the healthcare system or forgo treatment for financial reasons; they are a segment of the population that is statistically less healthy but generates lower costs. Learning from this flawed basis, the algorithm considered African Americans to be less sick and didn't recommend the appropriate treatment. When dealing with AI, the "trust but verify" principle must apply, always being able to trace the data back to its source.
Will AI allow us to be healthy to one hundred years old?
«Potentially yes. It is a necessary tool with great potential, but it must be used correctly."
First of all, we must overcome the poor quality and homogeneity of the data, which makes their use more difficult. Today there are many, but very different from each other. What is being done in this area?
«In fact it is a great challenge. We compare ourselves as a mass of very heterogeneous data. When a doctor fills out a chart he does so based on his needs. He doesn't have time to create codes for data processing. We are trying to overcome this problem with AI. Information is extracted from the free text of the medical record which is standardized so that it can be shared and compared. In this way the medical and healthcare staff does not change the way they compile the records and does not have an administrative burden. AI collects information to create structured data that contributes to diagnosis, research and effective treatments.
Going further, AI could help fill out medical records, with a final check, a validation, by the doctor. This is a key point that must always be reiterated: AI has superior capabilities compared to a single individual, but it should not replace them, but rather help make their work easier and more efficient.
What projects are you working on at the SUPSI MeDiTech Institute?
«An important line of research concerns the analysis of unstructured data: biomedical signals such as the electrocardiogram, the electroencephalogram, the characteristics of sleep; measurable data on human beings. We also deal with image analysis and information extraction from textual content present in the electronic medical record or in other biomedical texts, such as clinical research protocols. With this data we develop algorithms capable of predicting events.
We have numerous projects underway. The analysis of ECGs with AI has allowed us to identify arrhythmias imperceptible to the human eye, which can be traced back to rare heart diseases such as Brugada syndrome. Another use involves analyzing thermal images of the breast to increase the identification of breast cancer. We develop sensors and devices capable of detecting viruses in the air even at very low concentrations. Finally, we want to use data collected from wearable devices to monitor stress levels; or create systems to identify patients eligible for clinical trials through their medical records."
He has been directing MeDiTech for a few months after a long experience abroad, many years in the United States followed by an experience in the Netherlands. What strategic direction do you want to give to the institute?
«I certainly want to further develop connections with the cantonal, national and European healthcare sector to find the right synthesis between medical needs and our engineering and IT capabilities. A second aspect concerns responsible AI: we want to create reliable, bias-free and transparent algorithms. Finally, make a contribution to precision health with apps, wearable devices and integrated systems that collect biomedical signals even outside hospitals to allow us to offer a tailor-made approach to each individual based on their specificities."
