Pattern recognition

09/13/2017

Pattern recognition in clinical data

Pattern recognition is the process of classifying input data into objects or classes based on key features. Pattern recognition are methods that can be used for object detection and, classification of e.g. patients which are, per definition, responders to a certain treatment. The pattern in data before treatment starts are revealed by using different statistical techniques. E.g. Multiple logistic regression shows that the responders to treatment A shows a pattern of high baseline value of variable X1, low value of X2 and low value of X3. A classification rule can be made and used to classify patients to groups Responder or Non-responder, and accuarcy, sensitivity and specificity can be calculated, indicating the utility of the classification rule. A probability of the subject belonging to the group of interest can be estimated. If there are many subjects, a training set can be used for creating the classification rule which then is tested on the test set. Other techniques to calculate accuracy are more applicable when having smaller datasets, e.g. leave-"n"-out. Other areas than responder classification are diagnostization, genomic stratification etc.

Hans Fagertun has wide experience in pattern recognition methods and can in collaboration with you/your team, define areas to investigate. This includes defining subgroups of patients with specific characteristics that will give additional information to the public about treatments you offer.