IP Library › Granted Patent US 12,224,062
Granted Patent B2
US 12,224,062 · App. 17/164,433 · Granted Feb 11, 2025

Utilizing neural network models for recommending and adapting treatments for users

Inventors: Gaston Besanson (Barcelona, ES); Frode Huse Gjendem (Barcelona, ES); Bernabé Marcos Montes (Barcelona, ES); Joan Verdu Arnal (Taragona, ES)
Assignee: Accenture Global Solutions Limited
G16H50/20A61B5/4836A61B5/7267G06N3/045G06N3/08G16H10/60G16H20/10G16H50/30G16H50/70
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,224,062
App. No.
17/164,433
Granted
Feb 11, 2025
Kind
B2
Abstract

A device may receive user data identifying vitals of users when receiving treatments and dosages of the treatments, and may process the user data, with a divergence model, to determine divergence data identifying divergences between the users. The device may process the divergence data, with a clustering model, to group the users into clusters of users, and may train a first neural network model, with the user data, to generate a trained first neural network model. The device may train a second neural network model, with the user data, to generate a trained second neural network model, and may generate a treatment model based on the trained first and second neural network models. The device may process new user data identifying a new user, with the treatment model, to determine a recommended treatment for the new user, and may perform one or more actions based on the recommended treatment.

Claims (112)

1. A method, comprising:

receiving, by a device, user data identifying vitals of users when receiving treatments and

dosages of the treatments;

processing, by the device, the divergence data, with a clustering model, to group the users into clusters of users, wherein the clustering model includes a hierarchical clustering model and comprises applying the hierarchical clustering model to the divergence data to group the users into the clusters of users;

training, by the device, a first neural network model, with the user data, to identify treatments for the clusters of users and to generate a trained first neural network model, wherein the training of the first neural network model includes:

receiving, by the device, a set of observations;

determining, by the device, a target variable for the set of observations; and

training, by the device, the first neural network model using the target variable;

training, by the device, a second neural network model, with the user data, to determine simulated outcomes for the treatments and to generate a trained second neural network model, wherein the training of the second neural network model includes:

obtaining, by the device, additional user data;

determining, by the device, the simulation of outcomes based on the additional user data; and

causing, by the device, the second neural network model to train, based on the simulated outcome;

generating, by the device, a treatment model based on the trained first neural network model and the trained second neural network model;

training, by the device, the treatment model with the additional user data to generate a trained treatment model;

processing, by the device, new user data identifying a new user, with the trained treatment model, to determine a recommended treatment for the new user; and

performing, by the device, one or more actions based on the recommended treatment for the new user.

2. The method of claim 1 , wherein receiving the user data comprises:

receiving historical user data identifying historical vitals of the users when receiving the treatments and historical dosages of the treatments; and

receiving simulated user data identifying simulated vitals of the users when receiving the treatments and simulated dosages of the treatments.

3. The method of claim 1 , further comprising:

processing the new user data, with the divergence model, to determine new divergence data identifying a divergence between the clusters of users and the new user; and

processing the new divergence data, with the clustering model, to assign the new user to one of the clusters of users prior to processing the new user data with the trained treatment model.

4. The method of claim 1 , wherein processing the user data, with the divergence model, to determine the divergence data identifying the divergences between the users comprises:

calculating pairwise Jensen-Shannon divergences between the users; and

generating a distance matrix based on the pairwise Jensen-Shannon divergences between the users,

wherein the distance matrix corresponds to the divergence data.

5. The method of claim 1 , wherein training the first neural network model, with the user data, to identify the treatments for the clusters of users and to generate the trained first neural network model comprises:

calculating variances associated with the treatments for the clusters of users; and

generating the trained first neural network model when the variances satisfy a threshold variance.

6. The method of claim 1 , wherein training the second neural network model, with the user data, to determine the outcomes for the treatments and to generate the trained second neural network model comprises:

calculating variances associated with the outcomes for the treatments; and

generating the trained second neural network model when the variances satisfy a threshold variance.

7. A device, comprising:

one or more memories; and

one or more processors, communicatively coupled to the one or more memories, configured to:

receive user data identifying vitals of users when receiving treatments and dosages of the treatments,

wherein the user data includes one of:

historical user data identifying historical vitals of the users when

receiving the treatments and historical dosages of the treatments, or

simulated user data identifying simulated vitals of the users when

receiving the treatments and simulated dosages of the treatments;

process the user data, with a divergence model, to determine divergence data identifying divergences between the users, wherein the divergence model uses a kernel density estimation (KDE) and a Jensen-Shannon method to determine divergences between the users;

process the divergence data, with a clustering model, to group the users into clusters of users, wherein the clustering model includes a hierarchical clustering model and comprises applying the hierarchical clustering model to the divergence data to group the users into the clusters of users;

train a first neural network model, with the user data, to identify treatments for the clusters of users and to generate a trained first neural network model, wherein the training of the first neural network model includes:

receiving, by the device, a set of observations;

determining, by the device, a target variable for the set of observations; and

training, by the device, the first neural network model using the target variable;

train a second neural network model, with the user data, to determine simulated outcomes for the treatments and to generate a trained second neural network model, wherein the training of the second neural network includes:

obtaining, by the device, additional user data;

determining, by the device, the simulation of outcomes based on the additional user data; and

causing, by the device, the second neural network model to train, based on the simulated outcome;

generate a treatment model based on the trained first neural network model and the trained second neural network model;

train the treatment model with the additional user data to generate a trained treatment model;

process new user data identifying a new user, with the trained treatment model, to determine a recommended treatment for the new user; and

perform one or more actions based on the recommended treatment for the new user.

8. The device of claim 7 , wherein the one or more processors are further configured to:

cause the recommended treatment to be implemented for the new user;

receive outcome information identifying outcomes associated with implementing the recommended treatment for the new user; and

adapt the trained treatment model for the new user based on the outcome information.

9. The device of claim 8 , wherein the one or more processors are further configured to:

determine an updated recommended treatment for the new user based on adapting the trained treatment model for the new user; and

cause the updated recommended treatment to be implemented for the new user.

10. The device of claim 7 , wherein the one or more processors, when performing the one or more actions based on the recommended treatment, are configured to:

cause the new user to receive the recommended treatment;

receive feedback associated with the new user receiving the recommended treatment; and

update the trained treatment model based on the feedback.

11. The device of claim 7 , wherein the one or more processors, when performing the one or more actions based on the recommended treatment, are configured to:

define a range of treatments for the new user based on the recommended treatment and based on a risk threshold; and

cause the new user to receive the range of treatments.

12. The device of claim 7 , wherein the one or more processors, when performing the one or more actions based on the recommended treatment, are configured to:

cause the new user to receive the recommended treatment;

monitor vitals of the new user when receiving the recommended treatment; and

update the recommended treatment based on the vitals.

13. The device of claim 7 , wherein the one or more processors, when performing the one or more actions based on the recommended treatment, are configured to:

retrain one or more of the divergence model, the clustering model, the first neural network model, or the second neural network model based on the recommended treatment,

wherein the first neural network model and the second neural network model implement a reinforcement learning model.

14. A non-transitory computer-readable medium storing a set of instructions, the set of instructions comprising:

one or more instructions that, when executed by one or more processors of a device, cause the device to:

receive user data identifying vitals of users when receiving treatments and dosages of the treatments;

process the user data, with a divergence model, to determine divergence data identifying divergences between the users, wherein the divergence model uses a kernel density estimation (KDE) and a Jensen-Shannon method to determine divergences between the users;

process the divergence data, with a clustering model, to group the users into clusters of users, wherein the clustering model includes a hierarchical clustering model and comprises applying the hierarchical clustering model to the divergence data to group the users into the clusters of users;

train a first neural network model, with the user data, to identify treatments for the clusters of users and to generate a trained first neural network model, wherein the training of the first neural network model includes:

receiving, by the device, a set of observations;

determining, by the device, a target variable for the set of observations; and

training, by the device, the first neural network model using the target variable;

train a second neural network model, with the user data, to determine simulated outcomes for the treatments and to generate a trained second neural network model, wherein the training of the second neural network model includes:

obtaining, by the device, additional user data;

determining, by the device, the simulation of outcomes based on the additional user data; and

causing, by the device, the second neural network model to train, based on the simulated outcome;

generate a treatment model based on the trained first neural network model and the trained second neural network model;

train the treatment model with the additional user data to generate a trained treatment model;

process new user data identifying a new user, with the divergence model, to determine new divergence data identifying a divergence between the clusters of users and the new user;

process the new divergence data, with the clustering model, to assign the new user to one of the clusters of users;

process the new user data, with the trained treatment model, to determine a recommended treatment for the new user based on the one of the clusters of users assigned to the new user; and

perform one or more actions based on the recommended treatment for the new user.

15. The non-transitory computer-readable medium of claim 14 , wherein the one or more instructions, that cause the device to process the user data, with the divergence model, to determine the divergence data identifying the divergences between the users, cause the device to:

calculate pairwise Jensen-Shannon divergences between the users; and

generate a distance matrix based on the pairwise Jensen-Shannon divergences between the users,

wherein the distance matrix corresponds to the divergence data.

16. The non-transitory computer-readable medium of claim 14 , wherein the one or more instructions, that cause the device to train the first neural network model, with the user data, to identify the treatments for the clusters of users and to generate the trained first neural network model, cause the device to:

calculate variances associated with the treatments for the clusters of users; and

generate the trained first neural network model when the variances satisfy a threshold

variance.

17. The non-transitory computer-readable medium of claim 14 , wherein the one or more instructions, that cause the device to train the second neural network model, with the user data, to determine the outcomes for the treatments and to generate the trained second neural network model, cause the device to:

calculate variances associated with the outcomes for the treatments; and

generate the trained second neural network model when the variances satisfy a threshold variance.

18. The non-transitory computer-readable medium of claim 14 , wherein the one or more instructions further cause the device to:

cause the recommended treatment to be implemented for the new user;

receive outcome information identifying outcomes associated with implementing the recommended treatment for the new user;

adapt the trained treatment model for the new user based on the outcome information;

determine an updated recommended treatment for the new user based on adapting the trained treatment model for the new user; and

cause the updated recommended treatment to be implemented for the new user.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Feb 1, 2021
From: BESANSON, GASTON; GJENDEM, FRODE HUSE; MARCOS MONTES, BERNABÉ; VERDU ARNAL, JOAN
To: ACCENTURE GLOBAL SOLUTIONS LIMITED
Reel/Frame 055102/0149 →
Priority Claims (1)
EP 20383037 · Nov 30, 2020 · regional
Continuity (1)
Related Publication 20220172838A1 · Jun 2, 2022
References Cited (39)
US 20140279746A1 · De Bruin · 2014 [cited by examiner]
US 20190267121A1 · de Sousa Moura · 2019 [cited by examiner]
US 20200118458A1 · Shriberg · 2020 [cited by examiner]
US 20200143922A1 · Chekroud · 2020 [cited by examiner]
US 20210117842A1 · Smith · 2021 [cited by examiner]
US 20210202088A1 · Neumann · 2021 [cited by examiner]
US 20210232934A1 · Lai · 2021 [cited by examiner]
US 20210265064A1 · Hendriks · 2021 [cited by examiner]
US 20220344049A1 · Hall · 2022 [cited by examiner]
WO WO2019174898A1 · 2019 [cited by examiner]
Zarei, Anahita, et al. “An intelligent system for prediction of orthodontic treatment outcome.” The 2006 IEEE International Joint Conference on Neural Network Proceedings. IEEE, 2006. (Year: 2006). [cited by examiner]
Sinha, Yash Pratyush, et al. “Contextual care protocol using neural networks and decision trees.” 2018 Second International Conference on Advances in Electronics, Computers and Communications (ICAECC). IEEE, 2018. (Year… [cited by examiner]
Allam, Ahmed, et al. “Patient similarity analysis with longitudinal health data.” arXiv preprint arXiv:2005.06630 (May 2020). (Year: 2020). [cited by examiner]
Hatwell, Julian, Mohamed Medhat Gaber, and R. Muhammad Atif Azad. “Ada-WHIPS: explaining AdaBoost classification with applications in the health sciences.” (Oct. 2020). (Year: 2020). [cited by examiner]
Doctor, Faiyaz, Raouf NG Naguib, and Rahat Iqbal. “A Neuro-Fuzzy Approach for Identifying Practice Variations Based on Modelling Relationships between Clinical Variables and Treatment Decisions.” 2011 Developments in E-… [cited by examiner]
Johnson, Alistair EW, et al. “Machine learning and decision support in critical care.” Proceedings of the IEEE 104.2 (2016): 444-466. (Year: 2016). [cited by examiner]
Singh, Chandan, W. James Murdoch, and Bin Yu. “Hierarchical interpretations for neural network predictions.” arXiv preprint arXiv: 1806.05337 v2 (2019). (Year: 2019). [cited by examiner]
Dalla Man et al., “The UVA/PADOVA Type 1 Diabetes Simulator: New Features,” Journal of Diabetes Science and Technology, 2014, vol. 8(1), pp. 26-34 [retrieved from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4454102]. [cited by applicant]
Clarke et al., “Statistical Tools to Analyze Continuous Glucose Monitor Data,” Diabetes Technology & Therapeutics, 2009, vol. 11, Supplement 1, pp. 45-54 [retrieved from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2903… [cited by applicant]
Xie, simglucose, (2021), GitHub repository [retrieved from https://github.com/jxx123/simglucose]. [cited by applicant]
De Paula, et al., “On-line policy learning and adaptation for real-time personalization of an artificial pancreas,” Expert Systems with Applications, vol. 42, Issue 4, Mar. 2015, pp. 2234-2255. [cited by applicant]
Yao et al., “Direct Policy Transfer via Hidden Parameter Markov Decision Processes,” The 2nd Lifelong Learning: A Reinforcement Learning Approach (LLARLA) Workshop, Stockholm, Sweden, FAIM 2018, 7 pages. [cited by applicant]
Parbhoo et al., “Combining Kernel and Model Based Learning for HIV Therapy Selection,” AMIA Joint Summits on Translational Science Proceedings, 2017, pp. 239-248. [cited by applicant]
Fox et al., “Reinforcement Learning for Blood Glucose Control: Challenges and Opportunities,” Reinforcement Learning for Real Life (RL4RealLife) Workshop in the 36th International Conference on Machine Learning, Long Be… [cited by applicant]
Luckett et al., “Estimating Dynamic Treatment Regimes in Mobile Health Using V-learning,” 2017, 26 pages [retrieved from http://arxiv.org/abs/1611.03531v2]. [cited by applicant]
Killian et al., “Robust and Efficient Transfer Learning with Hidden Parameter Markov Decision Processes,” 31st Conference on Neural Information Processing Systems (NIPS 2017), Long Beach, CA, 12 pages. [cited by applicant]
Tordesillas et al., “Personalized Cancer Chemotherapy Schedule: a numerical comparison of performance and robustness in model-based and model-free scheduling methodologies,” 2019, 8 pages, [retrieved from https://arxiv.… [cited by applicant]
Ahn et al., “Drug scheduling of cancer chemotherapy based on natural actor-critic approach,” Biosystems, vol. 106, Issues 2-3, 2011, pp. 121-129. [cited by applicant]
Daskalaki et al., “Personalized tuning of a reinforcement learning control algorithm for glucose regulation,” 2013 35th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 201… [cited by applicant]
Jalalimanesh et al., “Simulation-based optimization of radiotherapy: Agent-based modeling and reinforcement learning,” Mathematics and Computers in Simulation, vol. 133, Mar. 2017, pp. 235-248. [cited by applicant]
Ngo et. al., “Reinforcement-learning optimal control for type-1 diabetes,” 2018 IEEE EMBS International Conference on Biomedical & Health Informatics (BHI), 2018. [cited by applicant]
Padmanabhan et al., “Reinforcement learning-based control of drug dosing for cancer chemotherapy treatment,” Math Biosciences, vol. 293, Nov. 2017, pp. 11-20. [cited by applicant]
Hassani et al., “Reinforcement Learning Based Control of Tumor Growth with Chemotherapy,” 2010 International Coriference on System Science and Engineering, Aug. 2010, pp. 185-189. [cited by applicant]
Liu et al., “Deep Reinforcement Learning for Dynamic Treatment Regimes on Medical Registry Data,” Healthc Inform., Aug. 2017, pp. 380-385. [cited by applicant]
Cruz-Lopez, “Deep Reinforcement Learning based Insulin Controller for Effective Type-1 Diabetic Care,” MConf2019, New York City, [retrieved from https://mlconf.com/sessions/deep-reinforcement-learning-based-insulin-cont… [cited by applicant]
“What is Continuous Glucose Monitoring (CGM)?” DEXCOPM Continuous Glucose Monitoring, 2021, 11 pages [retrieved from https://www.dexcom.com/continuous-glucose-monitoring on May 14, 2021]. [cited by applicant]
Anonymous: “AI and Personalised Medicine Artificial Intelligence Technology, Media & Telecommunications United Kingdom International Law Finn CMS”, Oct. 2, 2019 (Oct. 2, 2019), pp. 1-4, XP055805961, Retrieved from the I… [cited by applicant]
Byrne S., et al., “Using Neural Nets for Decision Support in Prescription and Outcome Prediction in Anticoagulation Drug Therapy”, Proceeding of Fifth International Workshop on Intelligent Data Analysis in Medicine and … [cited by applicant]
Extended European Search Report for Application No. EP20383037.7, mailed on Jun. 1, 2021, 10 pages. [cited by applicant]