Distance/divergence between graphs towards personalised medicine
Physicians often seek a non-invasive and reliable, real-valued score of the severity with which a patient is suffering from a disease, where said score embodies information on multiple clinical markers that the patient manifests.
We offer a method for learning a relative score of a patient, as a statistical distance or divergence between the probability of a random graph variable learnt given the data on such clinical markers relevant for one patient, and that given another, where the datasets are disparately long, given differential patient survivability.
In this ongoing project, such scoring has been undertaken for patients suffering from an Oncological disease.
Such an inter-graph distance/divergence has also been used to compute the recovery trajectory of a patient who is enrolled in a physical rehabilitation programme to regain their mobility that was impeded following an injury or critical illness.
The time series on the locations of joints in the patient’s skeletal framework is recorded an a virtual platform at every instance of the patient undertaking an exercise, and the distance/divergence between the probability of a random graph variable that is learnt given such time series data generated on one instance of undertaking this exercise, and that generated on the previous instance.
Our planned next step is the prediction of the recovery trajectory of a prospective patient.