Censor (Master thesis, SDU): INTELLIGENT DATA ANALYSIS FOR MEDICAL DIAGNOSIS BY USING MICROSOFT MACHINE LEARNING

Aktivitet: Ekstern undervisning og EksamenEksamination

Se relationer på Aarhus Universitet

Michael Alrøe - Eksaminator

Jennifer Faridi - Eksaminator

Throughout the health-care sector, patient journal notes are the principal practice of communicating across health-care professionals through a patients' entire treatment. The evaluation of patient journals is a manual process where the responsibility of capturing all crucial information lies with the person reviewing the journal. The consequence of missing any crucial information could lead to a misdiagnosis or wrong dosage. In the United States, physicians spend 2 hours on reading and writing patient journals for every 1 hour they spend with a patient.
The annual cost of physicians spending time on patient journals is over $365
billion, which is more than the U.S. spends on treating any dominant class of
disease.
This project aims to create an intelligent data analysis for medical patient
journals by using natural language processing methods. Information extraction
methods were used to provide an overview of the journals to improve efficiency
and the ability to evaluate the journals. Methods such as topic modelling, n-
gram and bag of words were used. The algorithm will be trained in a data
science virtual machine, where the code will be based on unsupervised learning,
as the journal data was not labelled.
The results from the algorithm were visualized in an application, where a
physician from Herlev Hospital evaluated the output. It was stated that the
model needed to be trained on more data, as only 5 out of 8 topics from the
topic model were possible to interpret. However, the results from the n-gram
and bag of words were qualified to be used in a real-life setting.
jun. 2019

ID: 159857786