Integrated system and method for personalized stratification and prediction of neurodegenerative disease

1. A method for predicting a disease state and/or progression rate performed on a computing device having a processor, memory, and one or more code sets stored in the memory and executed in the processor, the method comprising:

receiving feature-based representations of a first plurality of patients, the feature-based representations including static and dynamic features;

assigning a second plurality of plurality of patients among a plurality of patient clusters, each of the plurality of second patients represented with said static and dynamic features;

training a cluster-specific classifier for each of the plurality of patient clusters, each respective cluster-specific classifier operative to predict a disease state for a portion of the second plurality of patients of a respective one of the plurality of patient clusters;

iteratively running a training cycle until a performance measure is achieved, the training cycle including:

predicting disease states for the portion of the second plurality of patients with each of the respective cluster-specific classifiers,

reassigning at least one of the plurality of second patients to a different one of the plurality of patient clusters, wherein the respective cluster-specific classifier of the different patient cluster predicts a disease state and/or progression rate best matching a true disease state and/or a progression rate of said at least one of the plurality of second patients, and

retraining each of the respective cluster-specific classifiers using patients associated with each of the plurality of patient clusters after the reassigning;

receiving feature-based representations of a new patient;

assigning the new patient to a particular one of the plurality of patient clusters based on the received feature-based representations of the new patient; and

using the retrained cluster-specific classifier of the particular patient cluster to predict a disease state and/or progression rate of the new patient.

2. The method of claim 1 , wherein said receiving the feature-based representations of the first plurality of patients comprises receiving a plurality of observed features for each of the plurality of first patients; and

wherein said feature-based patient representations for each said of the plurality of first patients is based on a combinatorial combination of said observed features.

3. The method of claim 1 , assigning said new patient to the particular patient cluster in accordance with a cluster-specific prediction most closely matching an approximate disease state and/or approximate progression rate determined using a general non-cluster prediction model.

4. The method according to claim 3 , comprising training said general non-cluster prediction model using said received feature-based representation of patients; wherein said assigning of said new patient comprises:

predicting an approximate disease state and/or approximate progression rate for said new patient using said general non-cluster prediction model;

predicting a plurality of cluster-specific disease states and/or cluster-specific progression rates for said new patient, a cluster-specific disease state and/or cluster-specific progression rate predicted using each of the respective cluster-specific classifiers; and

assigning said new patient to a cluster a cluster-specific disease state and/or cluster-specific progression rate best matches said approximate disease state and/or approximate progression rate;

outputting said disease state or and/progression rate predicted by said assigned cluster-specific classifier for said new patient.

5. The method according to claim 1 , wherein said receiving the feature-based representations of the first plurality of patients comprises receiving feature-based representations of neurodegenerative disease (ND) patients; and

wherein said disease state and/or progression rate are characterized by Functionality Measure (FM) items and/or their change.

6. The method according to claim 1 , wherein said dynamic features include multiple values over time.

7. The method according to claim 1 , wherein said performance measure is reached when less than a selected percentage of said second plurality of patients are reassigned.

8. The method according to claim 7 , wherein said percentage is 5%.

9. The method according to claim 1 , wherein said training is intended for underfitting the training data of each of the plurality of patient clusters to provide a cluster-specific classifier which only fits a dominant group of patients in the each said patient cluster; and

wherein said iteratively running comprises decreasing underfitting of said retraining with each iteration.

10. The method according to claim 9 , wherein said training starts using one epoch; and

wherein said iteratively running comprises increasing a number of epochs with each iteration.

11. The method according to claim 1 , comprising receiving a pre-specified number of the plurality of patient clusters; and

wherein said assigning is among said number of the plurality of patient clusters.

12. The method according to claim 1 , wherein said predicting a disease state and/or progression rate includes providing a predicted disease state and/or progression rate at a future time.

13. The method according to claim 1 , wherein at least one of said cluster-specific classifiers is a temporal modelling based classifier.

14. The method according to claim 13 , wherein at least one of said cluster-specific classifiers is a long short-term memory (LSTM)-based cluster-specific classifier.

15. The method according to claim 1 , wherein said assigning comprises randomly assigning said second plurality of patients among said plurality of patient clusters.

16. The method according to claim 1 , wherein said assigning the second plurality of patients comprises, for said feature-based representations of patients:

calculating progression rates for individual features of said feature-based representations of patients;

for each of said second plurality of patients, establishing a multi-dimensional patient representation based on the feature progression rates to provide a plurality of multi-dimensional patient representations; and

clustering the plurality of multi-dimensional patient representations into a plurality of patient clusters.

17. The method according to claim 16 , wherein said establishing said multi-dimensional patient representation comprises:

grouping individual features into body function groups; and

calculating body function group progression rates for each of the body function groups.

18. A system for predicting a disease state and/or progression rate, the system comprising:

an input device operative to receive feature-based representation of a first plurality of patients, the feature-based representation including static and dynamic features;

a computing device configured to:

assign a second plurality of plurality of patients among a plurality of patient clusters, each of the plurality of second patients represented with said static and dynamic features,

train a cluster-specific classifier for each of the plurality of patient clusters, each respective cluster-specific classifier operative to predict a disease state for a portion of the second plurality of patients of a respective one of the plurality of associated patient clusters;

iteratively run a training cycle until a performance measure is achieved, the training cycle including:

predicting disease states for the portion of the second plurality of patients with each of the respective cluster-specific classifiers,

reassigning at least one of the plurality of second patients to a different one of the plurality of patient clusters, wherein the respective cluster-specific classifier of the different patient cluster predicts a disease state and/or progression rate best matching a true disease state and/or a progression rate of said at least one of the plurality of second patients, and

retraining each of the respective cluster-specific classifiers using patients associated with each of the plurality of patient clusters after the reassigning;

receive feature-based representations of a new patient;

assign the new patient to a particular one of the plurality of patient clusters based on the received feature-based representations of the new patient; and

use the retrained cluster-specific classifier of the particular patient cluster to predict a disease state and/or progression rate of the new patient; and

an output device operative to output the disease state and/or progression rate.

19. The system of claim 18 , wherein said input device is operative to receive a plurality of observed features for each of the plurality of first patients; and

wherein the feature-based patient representation for each of the plurality of first patients is based on a combinatorial combination of previously observed features.

20. The system of claim 18 , wherein the computing device is configured to assign said new patient to the particular patient cluster in accordance with a cluster-specific prediction most closely matching an approximate disease state and/or approximate progression rate determined using a general non-cluster prediction model.