Supervisor: Professor Ole Winther, DTU Compute.
Examiners:
Professor Jan Larsen, DTU Compute (Chairman).
Senior Lecturer MattiasOhlsson, Lund University
Professor Anders Søgaard, University of Copenhagen.
Chairperson at defence: Associate Professor Tobias Andersen, DTU Compute.
Abstract:
It is estimated that 6-8% of the European population will be affected by one of the known 7.000 rare diseases during their lifetime. Due to their rarity, diagnosis of rare diseases is often associated with yearlong diagnostic delays and errors. In other words, rare diseases (and their diagnosis) pose a huge societal problem.
FindZebra is an online search engine for rare diseases intended to act as a diagnosis decision support tool. Such online tools are increasingly being used by both doctors and patients
Due to the high number of persons suffering from a rare disease and due to the widespread use of online tools for diagnosis, even a modest increase in diagnostic performance of tools such as FindZebra can have a huge impact.
This thesis explore three successful directions for improving the performance of FindZebra as a diagnosis decision support tool. The first is by improving the retrieval accuracy by using a multitude of different methods such as neural diagnostic models, inclusion of structured data and corpus expansion. The second is by assisting the user following an unsuccessful search by a new method called Information Completion. The third direction is by providing native language support.
These improvements to the diagnostic process can result in faster diagnosis and fewer errors which in turn can help save lives, money and in general improve quality of life of persons suffering from rare diseases.
Further readings in DTU Orbit.
Everyone is welcome.