A new mathematical model for diagnosing IBD
Diagnosing IBD can be challenging and identifying the subtype (Crohn’s disease (CD) or ulcerative colitis (UC)) can be important when deciding appropriate treatments.
Using the endoscopy and histology data clinicians use on a every day basis to make the diagnosis, we trained a machine learning algorithm to identify patterns in the data that might help clinicians.
Our mathematical model identified patterns of inflammation in the gut within using anonymised data from 287 children with IBD and suggested the presence of a number of distinct sub-types of disease. Some of these groups included patients diagnosed with both CD and UC! This novel subgrouping was mainly influenced by different patterns of inflammation in the lower part of the gut and may be useful in deciding which information to use when trying to classify IBD.
Our machine learning model was able to predict whether a patient was affected by Crohn’s disease or ulcerative colitis with an 83% accuracy without knowing the doctor’s diagnosis – this is the highest accuracy so far reported for different models using clinical data. This mathematical model represents the starting point for a much complex modelling analysis which will include other types of data (e.g. genomics) with the hope of improving accuracy and clinical usefulness even more!
This work raised the interest of the scientific community and will be presented in Prague during the International Society of Computational Biology International Conference 2017.