Healthcare Process Modelling by Rule Based Networks
Han Liu First Year PhD Student
Alex Gegov, Jim Briggs, Mohammed BaderPhD Supervisors
Table of contents
• Health status monitoring• Treatment recommendation
Health Status Monitoring• A set of medical rules used to
predict health status is generated by a rule generation algorithm learning historical data and then converted into network structure illustrated in Figure 1
• Each node in input layer represents a medical feature
• Each node in middle layer represents a medical rule
• The output node represents the classification of health status, e.g. in risk or health
input conjunction output
Figure 1
If x1=1 and x2=1 then y=1
Treatment Recommendation 1. To classify patients into a particular category based on similarity
using K Nearest Neighbour.2. To retrieve treatments that have been applied to previous patients
classified into the same category as the current patient and find a list of candidate treatments by majority voting.
3. To classify these candidate treatments to one of rate scale of 1 to k and filter those treatments with negative classification.
4. To induce a list of association rules which have patient features on left hand side and medical features on right hand side and is represented by a network as illustrated in Figure 2.
5. To retrieve a list of most potential treatments that match the features represented by the right hand sides of association rules in order to recommend doctors a list of candidate choices.
If x1=1 and x2=1 then y1=1
Patient Features Medical Rules Medical Features
Figure 2
Thank you