Distorted picture
The AI model we’re developing compares personal data, such as exercise, eating habits, stress, and sleep patterns, from a smartwatch, for example, with that of a relevant population. Many existing models used in the medical profession give a biased picture because certain categories are over or under-represented in the data, or the data are incomplete. The data in these models often also ignore differences between groups in terms of age, gender, ethnicity, or susceptibility to certain diseases.
Predicting the probability of disease
TNO has developed methods to correct bias in models. This makes it possible to predict the risk of certain diseases, such as diabetes, much more accurately for certain groups of the population, and even for individuals. At the same time, highly personalised advice can be offered to each individual to prevent disease.
Doctors can use these models to compare their diagnosis of a patient with a relevant population. As new data are continually added to the self-learning system, the personalised, updated behavioural advice offered by the model will become increasingly sophisticated.
Personalised advice
How will this work in practice? Your age, gender, height, weight, and BMI will be known in the system. You’ll use your smartwatch or some other device to provide your daily data on exercise, food intake, heart rate, glucose levels, and more. The model will then supplement those data with data from a variety of sources and compare them with the population relevant to you.
This will generate a risk score for certain diseases, including an explanation of what the score is based on. The model will then tell you how much the current number of steps, calories burned, heart rate, and glucose level contribute to lowering or increasing this risk score and what you need to do that day to achieve your healthier lifestyle goals.
