Marco Spruit wants to develop a language model to improve healthcare
By making smart use of available data, the health and care of people can be substantially improved. Marco Spruit wants to use language and machine learning in the coming years to identify and solve the biggest care issues in the region of The Hague. He has been appointed Professor of Advanced Data Science in Population Health at the Leiden Institute of Advanced Computer Science (LIACS) and the Leiden University Medical Center (LUMC) on 1 December 2020.
Translational Data Science center
Spruit starts working on various projects from the LUMC-Campus The Hague. His main objective is to set up a Translational Data Science center. ‘I want to construct an authoritative language model for the Dutch healthcare sector,’ he says. ‘Analyzing Dutch language for the targeted improvement of health care is actually still in its infancy. The way a person talks says a lot about that person's mental state.’
Natural Language Processing
Ever since the 80's, AI researchers have been trying to understand human language skills by connecting the neural and logical approaches of Natural Language Processing. In recent years, the use of pre-trained contextual, deep-neural language models such as BERT/ClinicalBERT has taken off. These large-scale but generalized language models can then be better tuned to domain-specific language problems using Transfer Learning techniques.
The text analysis methods can also be extended with other sources containing structured data. Using Multimodal Machine Learning, for example, better risk profiling and welfare detection can be calculated by combining unstructured speech and text data with structured EHR data and sensor data from wearables.
Identifying psychological problems
By developing a good language model, Spruit expects, among other things, to be able to do a lot for vulnerable elderly people. For example, a language model can help to detect psychosocial complaints in this group. ‘Imagine: we are launching an application with which elderly people can enter into conversation. On the basis of these conversations we can find out if something is wrong. People with psychological problems such as borderline and elderly people with dementia, for example, often talk about themselves in the third person. By incorporating this into the model, we know when someone needs an intervention.’
Doctors and other healthcare professionals can also use data science to make their daily work easier. Machine learning offers many possibilities for this. This drives self-learning systems, which can improve models as more data is added. According to Spruit, machine learning is at the basis of designing and implementing an open, online, self-service, patient-sensitive platform for healthcare. ‘With the help of a smart medication review, physicians are able to, for example, easier monitor which medication is and which is not successful for patients. If you have a good overview of this, you can reduce incorrect medication and start different medication methods where necessary," explains Spruit.
Under the title Translational Data Science he is therefore investigating how Dutch healthcare processes can be improved reliably using data science technologies. According to him it's mainly about knowing what you're doing, and being able to interpret and reproduce results meaningfully. ‘Data can be interpreted in many different ways. That is why it is so important that learning models are set up with which we can interpret the data properly and take the right actions.’