Statistics based software system for finding the best treatment type and predicting the improvement.

摘要

#$%^&*AU2018100062A420180412.pdf#####ABSTRACT. This is an application software which allows users to make more accurate decisions when selecting the correct treatment type, correct dosage and correctly predicting recovery time based on past statistics. This system also let the patient to prioritize his own recovery requirements, (less painful, time taken etc.) and make treatment selection based on those requirements. Since the database is sharing globally users from multiple location can share their data and knowledge. They can make collaborative decisions when they change weights or add new factors or filter out some factors. This system basically running from two layers namely main grid spaces and child grid spaces. Since system may not have data to fill the full areas of both these layers, system use multi-variable gradient descend along with Monte Carlo Stimulation to fill these gabs. Gradient descend is running interactively between these two layers. When some part of empty space is filled in a one layer, it might automatically fill data in the other layer's empty spaces. Or it may generate data sufficient to establish minimum maximum data of the gradient descend of that layer. So this interactive data filling happens until system can no longer will empty spaces. After filling maximum empty spaces system use several ways of clustering methods to find values of the new data point. Here new patients record we consider as the new data point. Our goal is to find the best treatment, best dosage and expected recovery time for the new patient. Before establishing clusters, for each grid cell in the main grid space we find the best treatment type using the algorithm given above. Then grid cells with same best treatment type adjacent to each other and if they are continuous in any direction they come under same cluster. If the new point point is fallen inside one of these clusters, system use gradient descend to find the best treatment type and dosage. It uses the child grid space layer data of the poly data points of that cluster and make predictions. If those clusters does not hold sufficient data densities and no other best treatment type data in between clusters, system try to establish gradient descend maximum and minimum data along with other clusters and find values based on that gradient descend. If new data point is fallen outside of any of those clusters we use self tuning spectral clustering to build clusters. Here we could adjust number of clusters so that we could bring the new data point under one of those clusters. Or we could use gravitational attraction to find the best treatment type. To find the severity of symptoms we use severity index and to define how critical to the health of the patient this severity we use criticality index. To define minimum data requirements for each of these symptoms we use data density index. 1Background of the invention Here I am proposing an application software where system can use past statistical data to predict the best treatment type, expected improvements with each treatment type, the best dosages of drugs, and the best treatment combinations. This application software also generates a report which can be used to monitor a patent's improvement. This report predicts the improvements that are supposed to be achieved with the given treatment, and within a given time. Here our system uses a double layer system which comprises of a series of main multi-variable grids and a series of child multi-variable grids for each of those main grids, to make predictions. Here we use multi-variable gradient descend to find the data for unknown areas and use Monte Carlo stimulation to fill these empty spaces. Since we are storing all data in a centralized cloud based database, this database can be shared by multiple locations. Physicians can adjust weights collaboratively to make more accurate decisions. Goal of this application software is to make maximum use of available data and to make more accurate decisions, based on those data and shared knowledge. Alternative ways of achieving same functionality. Instead of multivariate gradient descend we could use cross-co-variance or cross-correlation interactively to generate same results. Other applications where we could use same concept. We could use the same concept in myriad of other applications where we need to make more accurate predictions based on available data. 2* Here weight change includes both severity detection weights and output dimentions weights. * Any weight changes requires most of the processed values to be 1 recalculated. Physician enter Systemn gather alll(2) Factor readings, - data relevant to diagnosis, seveity those symptoms Patient select - own output ye variables weight? Reading System find the No Compare physician most suitable correction Diagnosis with treatment & Needed minimu mYhits output and g requirede head Wegh offcoso Syste findingsTmpray djs hysicians Temporary adjust tput variable Evaluate the Collaboratively weights to find best Deviationdiagnosiso Try to match the treatment based on Deitio patient requirements Will changing,weight or adding a new feature wo k Yes Addaneror acor No If minimum Yes Either change erro datarcrdn required heads _4 Weight of factors or. errordataapproved add new factors Record adjustme-nts" & new factors values for future diagnosis Fig A01