Artificial intelligence (AI) models and applications to promote health care are being actively developed around the world. Researchers from the Faculty of Information Technology at the University of Jyväskylä, working with the Central Finland Health Care District are developing AI-based models to promote physicians’ work and the early diagnosis of knee osteoarthritis, write Sami Äyrämö and Juha Paloneva.

AI is proposed as a solution to a wide variety of problems. We agree that there has been significant progress in AI applications.

Originally computers were mainly capable of saving human labour by solving complex mathematical problems, whereas during the past decade they have started to learn, better and better, to tell in a human-like way what an image displays or to suggest how a sentence you have started could continue.

Many practical problems can already be solved or mitigated by means of AI, or it can aid researchers in seeking new insights from digital data.

Physicians’ experience of the signs of knee osteoarthritis can be passed on by means of AI

Our research collaboration relates to the care and diagnosis of a patient with knee pain at a health clinic.

Prolonged pain in the knee often derives from knee osteoarthritis, which may have different symptoms. Detection of knee osteoarthritis at an early stage is important in order to avoid unnecessary medical examinations and treatments. It can also help patients affect the development of the disorder by their own actions.

From an X-ray image, a physician can observe changes that indicate knee osteoarthritis.

A basic diagnostic method for knee osteoarthritis involves taking X-ray images of the knee. From an X-ray image, a physician can observe changes that indicate knee osteoarthritis. However, among people there is significant individual variation, and the interpretation of medical images is not always unambiguous. Especially detecting the signs of beginning knee osteoarthritis from images may be difficult even for a physician. Through specialisation and experience, however, a physician learns to make faster, more precise and consistent observations from the images.

AI can promote diagnostics by transferring an experienced physician’s expertise into an AI application, which can help a younger colleague’s decision-making by indicating, for instance, the significant points in an image that an experienced physician would look for.

This can help a less experienced physician notice the early, less obvious signs of the disorder among the normal variation appearing in the images. In such cases, it’s not the computer making the decisions, but it does assist the physician in decision-making.

Data plays a key role – medical experience “hidden” in X-ray image databases

Because the accumulated diagnostic experience of senior physicians cannot be programmed for computers in practice, we need a digital source, a database, of empirical knowledge to draw on. The data can be images, text, numerical values and so on.

Physicians can help us find relevant data sources for the development of AI, the data of which can then serve as raw material for AI developers.

Working together, the physicians and AI developers specify the objectives for a new AI application.

It then becomes the task of the AI researcher to develop and validate a model using algorithms and the data available, which would learn to reproduce, as precisely as possible, the diagnoses made by experienced physicians. Without relevant data and the contribution of physicians, an AI expert would be useless, and physicians, without an AI expert, would not have any AI models. There is strong interdependence between the two parties, so mutual trust and interaction are essential. Moreover, the data collected on people must be handled with the utmost care.

Some people also fear that AI will replace expert work. These fears are not realistic. Rather, letting AI handle the routine expert tasks would free the experts’ resources for more demanding tasks and for genuine contact with patients, for example. 

Long-term development of AI applications can bring advanced means for the promotion of wellbeing and health

It is not unusual that a good idea does not immediately lead to a working AI model. The reason for this is by no means in physicians’ inexperience nor in the algorithms’ incapability to create a working AI model from the data. If the data includes any consistent information, it can always be modelled – at least in theory.

A typical impediment remains, however: the qualitative and quantitative sufficiency of data. It is also a key limiting factor with regard to the advancement of AI models for early diagnostics in knee osteoarthritis.

Solutions for this problem are being sought by creating, for example, synthetic health data and extending expertise networks.

After the AI leaps and hype of the past ten years, we are entering a slightly slower developmental stage calling for patience, which will sooner or later be again followed by unexpected and perhaps greater advancements.

The multidisciplinary expertise and seamless cooperation between JYU and the Health Care District create excellent conditions in Central Finland for developing advanced means to improve wellbeing and health.

This science blog was written by Sami Äyrämö, Head of the Digital Health Intelligence Laboratory  from the Faculty of Information Technology, University of Jyväskylä, and Chief Medical Director, Professor Juha Paloneva from the Central Finland Health Care District. AI Hub Central Finland (AIHub) project is funded by ERDF European Regional Development Fund.

Get latest articles from The University of Jyväskylä’s stakeholder magazine into your email. You can cancel your subscription at any time.