Privacy-preserving computing for subject data used to develop artificial intelligence tools
Patent Information
- Application Number
- JP2023527687
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2020-11-10
- Filing Date
- 2021-11-08
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The challenge of developing artificial intelligence applications in healthcare is hindered by the difficulty in accessing and sharing sensitive patient data due to privacy and regulatory concerns, particularly with regulations like HIPAA, GxP, and GDPR, which complicates the training of effective machine learning models.
A method involving local and remote cloud servers for data processing, where the local server performs de-identification and anonymization of healthcare data within the same geographic region, and the remote server trains models using versioned datasets with expiration dates, ensuring compliance and privacy, while the trained models are deployed locally for analysis.
This approach enables the development and deployment of privacy-preserving AI tools by ensuring compliance with data regulations, protecting patient privacy, and reducing the time-to-market and complexity of model development.
Smart Images

Figure 00000000_0000_ABST
Abstract
Description
[Technical Field]
[0001] Priority Claim This application claims the benefit and priority of U.S. Provisional Patent Application No. 63 / 111,993, filed November 10, 2020, which is incorporated herein by reference in its entirety for all purposes.
[0002] Field of Invention This disclosure relates to digital and personalized healthcare, and in particular to technologies for privacy-preserving computing that protects the privacy of subjects while using subject data for secondary purposes such as training and deploying artificial intelligence tools. [Background technology]
[0003] Background of the Invention Modern computing paradigms, including cloud computing, data parallel cluster computing, and high-performance computing, combined with a wide variety of widely available machine learning and deep learning algorithm architectures, create an environment where a vast number of artificial intelligence (AI) applications can be developed to solve problems in almost any industry, provided that enough data is available to properly optimize the underlying algorithms. Clearly, access to data is a major barrier to the development of AI applications. In fact, many industries require the use of data from various sources to create robust, generalizable, and unbiased AI. A specific challenge is that data owners often cannot or will not share their data, or allow it to go beyond their control. This is understandable, as data often includes highly sensitive private and / or personal data and can be regulated in ways that make sharing difficult or impossible. These challenges are particularly difficult to overcome in the development of healthcare AI.
[0004] In healthcare AI, data-driven technological solutions are being developed to further personalize healthcare while reducing costs. Healthcare providers are innovating solutions to automate and streamline the process of analyzing patient data to determine medical predictions. Machine learning (ML) techniques can be used for many healthcare-related predictions, such as disease diagnosis and prognosis, as well as predicting treatment effectiveness. Because medical data typically includes the private / identifiable data of the subjects from which it is generated, compliance with government regulations (e.g., Health Insurance Portability and Accountability Act (HIPAA), Good Drug Handling Standards Quality Guidelines and Regulations (GxP), and General Data Protection Regulation (GDPR)) presents unique challenges for healthcare providers considering machine learning for medical analysis. Because training ML models can involve massive amounts of data, accessing enough data to deidentify and / or anonymize in order to train effective ML models can be difficult. Therefore, there is a need for advancements in corresponding software platforms built to provide accurate medical predictions while ensuring the confidentiality, availability, and integrity of protected healthcare information. [Overview of the project]
[0005] Summary of the Invention In various embodiments, a computer implementation method is provided, comprising: receiving subject data relating to a first subject from a first computing device associated with the first subject; performing a deidentification operation, an anonymization operation, or both, on the subject data to generate processed subject data; storing the processed subject data in a processed data store accessible to a local cloud server; sending batches of data to a remote cloud server, wherein the batches of data include processed subject data; receiving a generative model from the remote cloud server, wherein the generative model includes parameters partially derived from the processed subject data; receiving subsequent data relating to a second subject from a second computing device associated with the second subject; inputting the subsequent data into the generative model to analyze the subsequent data and generate inferences or predictions from the analysis of the subsequent data; and transmitting the inferences or predictions to the second computing device, the third computing device, or both for use in one or more operations performed by the second computing device, the third computing device, or a combination thereof.
[0006] In some embodiments, the local cloud server is physically located in the same geographical area as the subject.
[0007] In some embodiments, the same geographical area is the same country.
[0008] In some embodiments, subject data is healthcare data containing individually identifiable health information, and subsequent data is subsequent healthcare data containing individually identifiable health information.
[0009] In some embodiments, the same geographical region collectively shares a set of data regulations regarding the use and storage of individually identifiable health information.
[0010] In some embodiments, deidentification operations, anonymization operations, or both are performed on individually identifiable health information of subject data based on a set of data restrictions.
[0011] In some embodiments, the first computing device is the same as or different from the second computing device.
[0012] In some embodiments, the first computing device is a clinical device sensor, a handheld portable device, or a combination thereof.
[0013] In some embodiments, the second computing device is a clinical device sensor, a handheld portable device, or a combination thereof.
[0014] In some embodiments, the first subject is the same as or different from the second subject.
[0015] In some embodiments, the processed data store cannot access the remote cloud server.
[0016] In some embodiments, sending processed subject data to a remote cloud server as part of a data batch is done in response to the local cloud server not receiving a request to delete the processed subject data prior to sending the processed subject data.
[0017] In some embodiments, the method further includes storing subject data in a raw data store accessible to a local cloud server before performing a deidentification operation, anonymization operation, or both on the subject data; receiving a request from a remote cloud server to delete the subject data; and deleting the subject data from the raw data store in response to receiving the request to delete the subject data.
[0018] In some embodiments, processed subject data is not deleted from the processed data store.
[0019] In some embodiments, the transmission of processed subject data as part of a data batch is performed at a periodic or probabilistic timing such that the data batch includes data from multiple other subjects captured since the previous transmission of data to the remote cloud server.
[0020] In some embodiments, the inference or prediction is generated with respect to diagnosis, prognosis, treatment or therapy, identification of treatment or therapy protocols, detection or determination of disease status, identification or detection of biomarkers, reduction of non-compliance with treatment or therapy, reduction of operational costs, image analysis, marketing of treatment or therapy, automation of administrative tasks, assistance with medical procedures, or any combination thereof.
[0021] In some embodiments, one or more operations include communicating or displaying inferences or predictions, analyzing inferences or predictions, providing treatments or therapies, initiating treatment or therapy protocols, measuring biomarkers, providing notifications or reminders for treatments or therapies, acquiring healthcare data, reporting diagnoses or prognoses, analyzing images, providing marketing related to treatments or therapies, controlling medical devices or medical procedures, or any combination thereof.
[0022] In some embodiments, the method comprises sending a batch of subsequent data to a remote cloud server, the batch of subsequent data including inferences or predictions.
[0023] In some embodiments, the method further comprises performing a de-identification operation, an anonymization operation, or both, on subsequent data to generate processed subsequent data, storing the processed subsequent data in a processed data store accessible to a local cloud server, and sending a batch of data to a remote cloud server, the batch of data including the processed subsequent data.
[0024] In some embodiments, the method further comprises performing a de-identification operation, an anonymization operation, or both, on an inference or prediction to generate processed output data, storing the processed output data in a processed data store accessible to a local cloud server, and sending a batch of data to a remote cloud server, the batch of data including the processed output data.
[0025] In some embodiments, the method further comprises receiving, at a local cloud server, response data from a second computing device, a third computing device, or both, in response to receiving an inference or prediction in one or more operations or using an inference or prediction, performing a de-identification operation, an anonymization operation, or both, on the response data to generate processed response data, storing the processed response data in a processed data store, and sending a batch of data to a remote cloud server, the batch of data including the processed response data.
[0026] In some embodiments, the method further includes receiving a new generative model from a remote cloud server, wherein the new generative model includes parameters partially derived from processed response data; and replacing an existing generative model with the new generative model, wherein the replacement includes deleting the existing generative model from the local cloud server.
[0027] In various embodiments, a computer implementation method is provided, comprising: receiving processed subject data associated with a plurality of different subjects from a local cloud server, wherein the processed subject data is deidentified, anonymized, or both; associating the processed subject dataset with a versioned dataset; determining the expiration date of the versioned dataset; storing the versioned dataset in a versioned data store accessible to a remote cloud server, wherein the versioned dataset is stored in association with its expiration date; training a generative model using the versioned dataset; storing in the versioned data store an association between the versioned dataset and the generative model trained with the versioned dataset; and sending the generative model to the local cloud server for use in analyzing subsequent data and generating inferences or predictions from the analysis of the subsequent data.
[0028] In some embodiments, the local cloud server is physically located in the same geographical area as the subject.
[0029] In some embodiments, the same geographical area is the same country.
[0030] In some embodiments, the processed subject data is healthcare data containing individually identifiable health information, and the subsequent data is subsequent healthcare data containing individually identifiable health information.
[0031] In some embodiments, the remote cloud server is physically located in the same or a different geographical area as the local cloud server.
[0032] In some embodiments, the same or different geographical areas are the same or different countries.
[0033] In some embodiments, the version data store cannot access the local cloud server.
[0034] In some embodiments, the same geographical region collectively shares a set of data regulations regarding the use and storage of individually identifiable health information.
[0035] In some embodiments, individually identifiable health information from processed subject data is deidentified, anonymized, or both, based on a set of data regulations.
[0036] In some embodiments, the expiration date is determined based on the creation date of the versioned dataset, the date of receipt of the processed subject data, the expiration date of the informed consent document associated with the processed subject dataset, or any combination thereof.
[0037] In some embodiments, the method further includes storing generative models in a versioned datastore accessible from a remote cloud server, and deleting or removing the versioned datasets and generative models from the versioned datastore in response to the expiration of their validity period.
[0038] In some embodiments, the method further includes training another generative model using a versioned dataset and / or another versioned dataset; storing in a versioned datastore associations between the versioned dataset and / or another versioned dataset and another generative model trained on the versioned dataset and / or another versioned dataset; and sending the other generative model to a local cloud server to analyze subsequent data and generate inferences or predictions from the analysis of the subsequent data, wherein the other generative model generates inferences or predictions that are sent to the local cloud server with instructions to replace the generative model with the other generative model.
[0039] In some embodiments, the inference or prediction is generated with respect to diagnosis, prognosis, treatment or therapy, identification of treatment or therapy protocols, detection or determination of disease status, identification or detection of biomarkers, reduction of non-compliance with treatment or therapy, reduction of operational costs, image analysis, marketing of treatment or therapy, automation of administrative tasks, assistance with medical procedures, or any combination thereof.
[0040] Some embodiments of this disclosure include a system comprising one or more data processors. In some embodiments, the system includes a non-temporary computer-readable storage medium comprising instructions, wherein when the instructions are executed on one or more data processors, the one or more data processors cause one or more data processors to execute some or all of one or more of the methods and / or some or all of one or more processes disclosed herein. Some embodiments of this disclosure include a computer program product tangibly embodied in a non-temporary machine-readable storage medium comprising instructions configured to cause one or more data processors to execute some or all of one or more of the methods and / or some or all of one or more processes disclosed herein.
[0041] The terms and expressions used are for illustrative purposes only, not limitation, and in using such terms and expressions, there is no intention to exclude any equivalent or part of any of the features shown and described, however it should be recognized that various modifications are possible within the scope of the invention described in the claims. Accordingly, although the invention described in the claims is specifically disclosed by embodiments and optional features, modifications and variations of the concepts disclosed herein may be relied upon by those skilled in the art, and it should be understood that such modifications and variations are considered to be within the scope of the invention as defined by the appended claims. [Brief explanation of the drawing]
[0042] This disclosure is described in conjunction with the following attached drawings:
[0043] [Figure 1] This diagram illustrates a digital health platform that provides data-driven technical solutions across various embodiments.
[0044] [Figure 2] The diagrams show model systems representing various embodiments.
[0045] [Figure 3] This shows swimlane diagrams illustrating processes for privacy-preserving computing in digital health platforms, according to various embodiments.
[0046] [Figure 4] This flowchart illustrates privacy-preserving computing performed by a local cloud server on subject data used to develop artificial intelligence tools according to various embodiments.
[0047] [Figure 5]This flowchart illustrates privacy-preserving computing performed by a remote cloud server on subject data used to develop artificial intelligence tools according to various embodiments.
[0048] In the accompanying drawings, similar components and / or features may have the same reference label. Furthermore, various components of the same type may be distinguished by following the reference label with a dash and a second label to distinguish similar components. Where only the first reference label is used herein, the description is applicable to any similar component having the same first reference label, regardless of the second reference label. [Modes for carrying out the invention]
[0049] Detailed description of the invention I. Overview This disclosure describes technologies for privacy-preserving computing to protect the privacy of subjects while using subject data for secondary purposes such as training and deploying artificial intelligence tools. More specifically, embodiments of this disclosure provide digital and personalized healthcare platforms that facilitate the development and deployment of applications and / or algorithms in a privacy-preserving manner. While various embodiments of machine learning and algorithmic architectures in which AI and ML models (hereinafter simply referred to as models) are developed to solve problems in the healthcare industry are disclosed herein, it should be understood that these architectures and technologies may be implemented in other types of systems and settings. For example, these architectures and technologies may be implemented in the development of models in many industries (finance, life sciences, supply chain, national security, law enforcement, public safety, etc.) where the confidentiality of the data (e.g., whether including trade secrets or private data relating to individuals) prevents the sharing of data outside the boundaries of the organization responsible for its protection.
[0050] A key challenge when dealing with healthcare data is privacy. Healthcare data is provided in many different forms, including clinical data, clinical trial data, genomic data, subject-generated data, Internet of Things (IoT) data, social media data, registration data, survey data, and medical device data. Healthcare data typically includes protected health information, which is any information that may be transmitted or held by one of the entities covered under government regulations (e.g., HIPAA, GxP, and GDPR) that include personally identifiable health information. Personally identifiable health information includes any information that can identify a subject as an individual, such as name, address, date of birth, and social security number. It may also include any information relating to the subject's physical or mental health, the provision of health care to the individual, or payments for the provision of health care to the patient, whether current, past, or future. There are many legal, cultural, and technological challenges associated with collecting, sharing, and processing healthcare data for secondary purposes, such as training and deploying artificial intelligence tools. For example, many countries have strict privacy laws and regulations that must be followed when dealing with personally identifiable health information of individuals. This could make the collection, sharing, and processing of healthcare data a significant challenge.
[0051] To overcome these challenges, traditional technologies for privacy-preserving computing focus on protecting healthcare data and their associated models. This combination aims to provide a trustworthy, transparent, and auditable technical implementation (structured transparency) by ensuring control over input data and models, and guaranteeing the integrity of the computation process and its results. Such systems are built to withstand attacks on datasets, such as identity or membership inference / tracking (determining whether an individual exists in a given dataset) and feature / attribute re-derivation / re-identification (e.g., extraction of an individual's features from within a dataset by linkage attacks). The systems are also built to withstand attacks on algorithms or computation processes, such as alteration of algorithm parameters (e.g., by poisoning) or derivation of information about the dataset from them (model inversion / reconstruction). Finally, the systems are also configured to protect data and algorithms from theft both in storage and when transmitted over the network (asset / integrity protection).
[0052] Despite the aforementioned technologies for protecting healthcare data while performing collection, training, and deployment, many organizations avoid using healthcare data altogether due to the increasing time to market, cost, and complexity of model development. For example, the timeline required to complete the steps necessary to protect and use privacy-protected data for model development exceeds the typical return on R&D investment expectations.
[0053] To address these limitations and issues, the privacy-preserving computing techniques in this disclosure utilize optimized data flows that separate model training and deployment between local and remote servers. The local server is stored on a computing device (e.g., a cloud server) physically located within the region (e.g., a country or state) from which the data (e.g., healthcare data) originates. The remote server, on the other hand, is maintained regardless of the region from which the data originates. In some cases, the remote server is located in the same region as the local server (but isolated from each other). In other examples, the remote server is located in a different region from the local server. The local server is adapted to receive the data and perform deidentification and / or anonymization operations on the data to generate privacy-preserving data.
[0054] As used herein, “anonymization” is the act of permanently and completely removing personally identifiable information from data, such as by converting personally identifiable information into aggregated data. Anonymized data is data that can no longer be associated with an individual in any way. Once the elements that identify an individual are removed from this data, those elements cannot be reassociated with the data or the underlying individual. In contrast, as used herein, “deidentification” is the act of removing personally identifiable information in order to protect the privacy of an individual. Deidentified data does not necessarily have to be anonymized data. In other words, in the case of deidentified data, personally identifiable information may later be reassociated with the data or the underlying individual. The local server is then further adapted to send batches of data containing the privacy-protected data to a remote server for downstream processing.
[0055] The remote server is adapted to associate privacy-protected data with versioned datasets so that the privacy-protected data is subject to version control. The remote server is further adapted to determine the expiration date of the versioned dataset and to store versioned datasets with expiration dates in the data store. The remote server is further adapted to train a model using the versioned dataset. The model includes parameters learned or derived based on minimizing one or more loss functions. The remote server is further adapted to generate and store a mapping between the versioned dataset and the model in the data store so that the model is now connected to the versioned dataset and therefore also subject to version control. The remote server is further adapted to send the model to the local server for deployment. The model may then be implemented on the local server to be applied to data (e.g., providing analysis or predictions in response to processing healthcare data).
[0056] One exemplary embodiment of the present disclosure is a method performed by a local server, comprising: receiving subject data relating to a first subject from a first computing device associated with the first subject; performing a deidentification operation, an anonymization operation, or both, on the subject data to generate processed subject data; storing the processed subject data in a processed data store accessible to a local cloud server; sending batches of data to a remote cloud server, wherein the batches of data include processed subject data; receiving a generative model from the remote cloud server, wherein the generative model includes parameters partially derived from the processed subject data; receiving subsequent data relating to a second subject from a second computing device associated with the second subject; inputting the subsequent data into the generative model to analyze the subsequent data and generate inferences or predictions from the analysis of the subsequent data; and transmitting the inferences or predictions to the second computing device, the third computing device, or both for use in one or more operations performed by the second computing device, the third computing device, or a combination thereof.
[0057] Another exemplary embodiment of the present disclosure is a method performed by a remote server, which includes receiving processed subject data associated with multiple different subjects from a local cloud server, wherein the processed subject data is deidentified, anonymized, or both; associating the processed subject dataset with a versioned dataset; determining the expiration date of the versioned dataset; storing the versioned dataset in a versioned data store accessible to the remote cloud server, wherein the versioned dataset is stored in association with its expiration date; training a generative model using the versioned dataset; storing in the versioned data store an association between the versioned dataset and the generative model trained on the versioned dataset; and sending the generative model to the local cloud server for use in analyzing subsequent data and generating inferences or predictions from the analysis of subsequent data.
[0058] II. Digital Health Platform Figure 1 shows a simplified diagram of a digital health platform 100 for providing data-driven technical solutions according to various embodiments. In the illustrated embodiment, the digital health platform 100 includes client computing devices 105 connected to a cloud-based infrastructure 110 via a network 115 including a network gateway 120 and a network mesh 125. The infrastructure 110 is adapted to run services or software applications in service pods 130 using resources provisioned within a deployment ring 135 by a cloud service provider 140 (e.g., a distributed computing environment) using various hardware and cloud infrastructure (e.g., private or on-premises cloud infrastructure and public cloud infrastructure). These services or software applications may be delivered to users of the client computing devices 105 as web-based or cloud services, for example, under an AaaS or SaaS model. Several providers, such as Amazon, Google, and Oracle, offer cloud services. The term cloud service is generally used to refer to services made available to users on demand via a communication network such as the Internet by a service provider's system (e.g., infrastructure 110), such as a government regulatory entity. Therefore, consumers can use cloud services provided by service providers without having to purchase separate licenses, support, or hardware and software resources to support the services. For example, a cloud service provider's system can host one or more programs, and users can use one or more programs on demand via the internet without having to purchase infrastructure resources to run those programs.Cloud services are designed to provide easy and scalable access to applications, resources, and services.
[0059] In some cases, a user operating the client computing device 105 (e.g., a software or service consumer) utilizes one or more client applications to consume software products, services, or systems provided by various components 145 of the infrastructure 110. In other examples, a user operating the client computing device 105 (e.g., a developer) utilizes one or more client applications to upload the source code of software products, services, or systems provided by various components 145 of the infrastructure 110. Component 145 includes one or more processors, hardware components, or software components that can be run by a combination thereof. It should be understood that various different system configurations are possible and can differ from those shown for the digital health platform 100. Therefore, the embodiment shown in Figure 1 is an example of a distributed computing environment for implementing the digital health platform, and is not intended to be limiting.
[0060] Client computing devices 105 include various types of computing systems such as portable handheld devices, general-purpose computers such as personal computers and laptops, workstation computers, wearable devices, game systems, thin clients, various messaging devices, sensors or other sensing devices, etc. These computing devices support various mobile operating systems (e.g., Microsoft Windows Mobile). (登録商標), various types and versions of software applications and operating systems, including iOS (registered trademark), Windows Phone (registered trademark), Android (trademark), BlackBerry (registered trademark), Palm OS (registered trademark), for example, Microsoft Windows (登録商標) , Apple Macintosh (登録商標) , UNIX (登録商標) or UNIX-based operating systems, Linux or Linux-based operating systems such as Google Chrome (trademark) OS. Portable handheld devices can include mobile phones, smartphones (e.g., iPhone (登録商標) ), tablets (e.g., iPad (登録商標) ), personal digital assistants (PDAs), etc. Wearable devices can include Fitbit Versa (商標) smartwatches, magic leap 1 (登録商標) and virtual reality (VR) or augmented reality (AR) systems such as Oculus (登録商標) , and other devices. Gaming systems can include various handheld gaming devices, Internet-connected gaming devices (e.g., Microsoft Xbox (登録商標) with or without gesture input devices, Sony PlayStation (registered trademark) systems, various gaming systems provided by Nintendo (registered trademark), and others). Client device 105 can run various different applications, such as various Internet-related applications, communication applications (e.g., email applications, short message service (SMS) applications), etc., and can use various communication protocols.
[0061] Network 115 is any type of network well known to those skilled in the art that can support data communication using any of the various available protocols, including but not limited to TCP / IP (Transmission Control Protocol / Internet Protocol), SNA (System Network Architecture), IPX (Internet Packet Switching), and AppleTalk®. Just as examples, Network 115 could be a local area network (LAN), Ethernet, Token Ring, wide area network (WAN), internet, virtual network, virtual private network (VPN), intranet, extranet, public switched telephone network (PSTN), infrared network, or wireless network (e.g., IEEE 1002.11 protocol suite, Bluetooth). (登録商標) This can be a network operating under any of the following and / or any other radio protocols, and / or any combination of these and / or other networks.
[0062] The network gateway 120 is a network node that forms a secure path between multiple networks 115 operating on the same or different protocols. The network gateway 120 can provide network security using one or more of the following technologies: a firewall to monitor incoming and outgoing network traffic, a virtual private network to provide a private and secure communication channel, security scanning to identify security vulnerabilities in the network, and an access manager for authentication and authorization services. The network gateway 120 routes network traffic using service connectors that manage access to routers and various software products, services, or systems (e.g., using a service subscription business model). The network mesh 125 is a local network topology in which infrastructure 110 (e.g., bridges, switches, and other infrastructure devices) connect directly, dynamically, and non-hierarchically to as many other nodes as possible, working together to efficiently route data between devices and nodes. The network mesh 125 manages connectivity using one or more of the following technologies: load balancing, product, service, or system discovery, network access, routing, and peering, traffic mirroring, etc. Network 115, network gateway 120, and network mesh 125 work together to manage all data flowing in or out of infrastructure 110.
[0063] Component 145 includes one or more general-purpose computers, dedicated server computers (including, for example, PC (personal computer) servers, purpose-specific servers, midrange servers, mainframe computers, rack-mount servers, etc.), server farms, server clusters, or any other suitable configuration and / or combination of computers or systems that operate individually or in combination to provide resources, data, services, or programs to client computing devices 105 via the network 115. Component 145 may further include other computing architectures that include virtualization, such as one or more virtual machines running a virtual operating system, or one or more flexible pools of logical storage devices that can be virtualized to maintain a virtual storage device. In various embodiments, component 145 is adapted to run one or more services or software applications that provide the functionality described in this disclosure.
[0064] Component 145 also includes one or more data repositories. These data repositories may be used to store data and other information in various embodiments. For example, one or more of the data repositories may be used to store information for providing data-driven technical solutions such as software as a medical device (SAMD), and for verifying and deploying source code for implementing the data-driven technical solution. The data repositories may reside in various locations. For example, a data repository used by a component may be local to the component, or it may be remote from the component and communicate with the component via a network-based or dedicated connection. The data repositories may be of different types. In certain embodiments, a data repository used by a component may be a database, such as a centralized database, a distributed database, a NoSQL database, or a relational database. One or more of these databases may be adapted to allow the storage, updating, and retrieval of data to and from the database in response to SQL-formatted commands. In certain embodiments, one or more of the data repositories may also be used by an application to store application data. The data repository used by the application may be of a different type, such as a key-value store repository, an object store repository, or a general storage repository supported by the file system.
[0065] Component 145 also includes compute nodes adapted to run one or more programs, such as services or software applications that provide the functionality described in this disclosure (e.g., services or software applications delivered as web-based or cloud services, or applications for implementing continuous integration and continuous deployment (CI / CD) systems). Each node is optionally a representation of a single machine implemented within a cluster of nodes. A single machine may be a physical machine (e.g., a server in a data center) or an Amazon Web Services server with a set of available CPU and RAM resources. (商標) These may be virtual machines hosted by a cloud provider such as AWS. In a cluster, nodes pool their resources to form a more powerful machine. Once one or more programs are deployed on the cluster, the cluster intelligently handles the distributed work to the individual nodes. When nodes are added or removed, the cluster can shift work as needed. Which individual machine is actually running the code is not important to one or more programs or to the infrastructure.
[0066] One or more programs deployed on one or more clusters are packaged as containers. Containers are a widely accepted standard, and various images can be defined for deploying one or more programs on infrastructure 110. Containerization allows infrastructure 110 to create self-contained execution environments. Any program and all its dependencies can be bundled into a single file and then shared on infrastructure 110. Container creation can be done programmatically, enabling a powerful, fully automated CI / CD pipeline used for verifying and deploying code on infrastructure 110. Containers are wrapped in a higher-level structure known as pods 130. Containers within the same pod 130 can share the same resources and local network. In some cases, containers can communicate with other containers within the same pod 130 as if they were on the same machine, while maintaining some degree of isolation from others. Pods 130 are used as replication units within infrastructure 110. If a program or resource is overloaded with processing and a single pod 130 instance cannot bear the load, infrastructure 110 may be configured to deploy new replicas of pod 130 to the cluster as needed. Even when there is no heavy load, it may be beneficial to have multiple copies of pod 130 running at all times in the generating system to enable load balancing and fault tolerance. One or more instances of pod 130 are provisioned in a cloud infrastructure system provided by one or more cloud service providers 140.
[0067] A cloud infrastructure system provided by one or more cloud service providers 140 includes infrastructure resources used to facilitate the provision of one or more instances of pods 130 that support various cloud services provided by infrastructure 110. To facilitate the efficient use of these resources for provisioning one or more instances of pods 130, the resources may be bundled into a set of resources or resource modules, "placement rings 135". Each resource module or placement ring 135 may contain a pre-integrated and optimized combination of one or more types of resources. In certain examples, different placement rings 135 may be pre-provisioned for different types of cloud services. For example, a first set of placement rings 135 may be provisioned for SAMD services, and a second set of placement rings 135 may contain a different combination of resources than the placement rings 135 in the first set of placement rings 135, and may be provisioned for data analytics services, and so on. For some cloud services, the resources allocated to provision the services may be shared among the services.
[0068] The digital health platform 100 further includes one or more kernels 150. Each kernel 150 is adapted to operate on each cloud infrastructure system provided by one or more cloud service providers 140. The kernel 150 is a cluster manager that provides resource allocation and isolation across distributed applications or frameworks throughout the digital health platform 100. The kernel 150 provides one or more programs with an application programming interface (API) for orchestrating services and software, including resource management and scheduling. The architecture of the kernel 150 includes agent nodes for executing tasks, master nodes for submitting tasks to the agent nodes, area managers for election and for looking up the addresses of the master nodes, and a framework for coordinating with the master nodes to schedule tasks on the agent nodes.
[0069] The digital health platform 100 further includes a CI / CD system 155. The CI / CD system 155 is implemented within a cloud infrastructure system, enabling the digital health platform 100 to frequently update, test, and deliver changes within the source code of software products, services, or systems. As detailed herein, in healthcare, there are government regulations regarding data security (e.g., data integrity and data privacy) that software must comply with. The CI / CD system 155 allows these policy regulations to be incorporated into the code, enabling compliance to be automatically tracked, verified, and reconfigured. In the SAMD example, data storage locations, server access control, and activity logging may be incorporated into the source code, ensuring user data is protected and managed during software use. Encryption and password-protected behavior may be further incorporated during continuous integration. During continuous delivery, security and monitoring tools can be used to track user activity and detect errors that could lead to security threats.
[0070] The CI / CD system 155 may also be used for provisioning models. While the model is initially trained using a dataset, over time, the model may drift or the data may change, requiring an updated model. If the model runs within a software application, the code associated with the software application can include triggers for when the model should be retrained. For example, the code could include instructions to retrain the model at predetermined time intervals when new training data becomes available or when it is determined that the model's performance falls below a threshold. Furthermore, software developers can explore variations in the model architecture and hyperparameters in a test environment based on monitoring the model's performance in the production environment or based on estimated improvements for model optimization. The CI / CD system 155 facilitates building, testing, and deployment to the production environment once it is determined that the model meets performance requirements.
[0071] III. Model System Figure 2 shows a simplified diagram of a model system 200 for privacy-preserving computing (including various components 145 of the infrastructure 110 described in relation to Figure 1) according to various embodiments. In the illustrated embodiments, the model system 200 includes a user 205 (e.g., a subject and / or operator), various devices 210 (e.g., a personal computer, medical device, IoT device, etc.), a local server 215, and a remote server 220. The local server 215 and the remote server 220 represent various scalable instances of the components of the infrastructure 110 described in relation to Figure 1. The local server 215 is maintained within the region (e.g., country or state) from which the data (e.g., healthcare data) originates. The remote server 220, on the other hand, is maintained anywhere, regardless of the region from which the data originates. In some cases, the remote server 220 is located in the same region as the local server 215 (but separated from each other). In other examples, the remote server 220 is located in a different region from the region of the local server 215. Although only single instances of local server 215 and remote server 220 are shown, multiple instances may be implemented within the digital health platform, and it should be understood that each local server 215 and remote server has its own set of components as shown in Figure 2.
[0072] Various devices 210 can be operated actively or passively by a user 205, in which case data can be generated and / or collected (for example, healthcare data may be generated and / or collected from a SAMD application running on a mobile device, or from a neuromodulatory device implanted in a subject). In some cases, a software development kit associated with one or more applications on device 210 is adapted to enable the collection and processing of data generated by one or more applications before transmitting the data to a local server 215 via secure communication. For example, the software development kit may provide a set of tools that enable data collection and deidentification. In other examples, raw data generated from device 210 or one or more applications on device 210 is transmitted to the local server 215 via secure communication without prior collection and / or processing (e.g., deidentification).
[0073] Device 210 can communicate securely with the digital health platform via the distributed computing environment connector and DNS (i.e., communicate in a manner less susceptible to eavesdropping or interception). DNS is a hierarchical distributed database that stores IP addresses and other data, enabling IP addresses to be looked up by name to forward calls to those IP addresses. For example, when the digital health platform receives a secure communication from device 210 for sending data, the distributed computing environment connector identifies an instance of local server 215 that can process the data, forwards the secure communication and the identifier of the instance of local server 215 that can process the data to DNS, which looks up the IP address and other data associated with the instance of local server 215 and forwards the secure communication to the gateway endpoint of the gateway associated with the instance of local server 215 (e.g., gateway 120 as described in relation to Figure 1).
[0074] Once connected to the gateway endpoint, device 210 can communicate securely with local server 215 and send data for storage and processing. In some cases, identity recognition can be used to monitor communication and data traffic with local server 215. This allows local server 215 to perform data access and auditing based on identity. Identity recognition maps user and device identifiers, allowing data transfer and access to be permitted or denied based on identity policies.
[0075] The local server 215 includes a data processor 225 adapted to deidentify, anonymize, store, and / or process data as part of one or more operations or services provided by the local server. Data received by the local server 215 may be stored according to one or more policies or rules governing the type of data received and the storage of the data (e.g., a set of data restrictions). For example, streaming data may be deidentified and stored in a raw data store. In contrast, batch and processed data may be anonymized and stored in a processed data store. Similarly, data received by the local server 215 may be processed according to one or more policies or rules governing the type of data received and the processing of the data (e.g., a set of data restrictions). For example, streaming data may be deidentified and processed according to one or more operations related to a streaming program (e.g., a DAG pipeline), and SAMD batch data may be anonymized and processed according to one or more operations related to a SAMD program (e.g., standardization of medical images for AI or ML model ingestion and deidentification of personal identifiers).
[0076] Deidentification involves one or more operations to remove or sever the link between direct and known indirect personal identifiers and real-world identities (personally identifiable information). One or more operations are reversible, and therefore the data can be re-identified. Anonymization involves one or more operations to remove or manipulate personally identifiable information, resulting in anonymized data that cannot be re-associated with any real-world identity. One or more operations for anonymization include data masking, pseudonymization, generalization, data swapping, data perturbation, synthetic data, or any combination thereof. Data masking is a technique to hide data whose values have been changed. For example, value characters in data may be replaced with symbols such as "*" or "x". Pseudonymization is a data management and deidentification technique that replaces private identifiers with false identifiers or pseudonyms, such as replacing the identifier "Bob Crown" with "John Duke". Generalization is a technique to remove parts of data or replace parts of data with common values. For example, generalization can be used to remove or replace all segments of social security numbers or health insurance numbers with the same sequence of numbers. Data swapping (also known as shuffling and substitution) is a technique used to rearrange dataset attribute values so that they do not match the original records. Data perturbation is a technique that slightly alters the original dataset by applying techniques such as rounding numbers and adding random noise. Synthetic data is a technique that algorithmically generates information that is not related to actual events. Synthetic data is used to create artificial datasets instead of altering the original dataset, or instead of using the original dataset as is and risking privacy and security.
[0077] The data processor 225 is further adapted to partition data and store it in data partitions containing similar data. Data may be stored together by the service / application / SAMD to which it is associated, device ID, country of origin, originating healthcare system, etc. The data processor 225 can securely communicate with the gateway endpoint of the gateway associated with the instance of the remote server 220 (e.g., gateway 120 as described in relation to Figure 1). Once connected to the gateway endpoint, the local server 215 can communicate securely with the remote server 220 and send data for further processing using extraction, transformation, and loading techniques via the application programming interface. The data received by the remote server 220 is stored in one or more data stores (e.g., local or remote data lakes or repositories to the remote server 220).
[0078] The remote server 220 comprises a training AI engine 230 adapted to train one or more algorithms and / or models 235 and generate a knowledge base 240 for a generative model 245. The training AI engine 230 may be adapted to run data stored in one or more data stores via the algorithms and / or models 235. In some embodiments, running includes executing a training workflow that includes creating one or more local instances of the algorithms and / or models 235; splitting the data into one or more sets of training data and one or more sets of test data; training one or more local instances of the algorithms and / or models 235 on the sets of training data to generate a generative model 245; optionally integrating the results from training each of the one or more local instances of the algorithms and / or models 235 into a fully federated generative model 245; running one or more sets of test data via the generative model 245; and calculating the performance of the generative model 245 based on the running of one or more sets of test data. In some embodiments, the execution includes performing a validation workflow that includes splitting, joining, and / or transforming the data into one or more validation datasets, running one or more validation datasets through the generative model 245, and calculating the performance of the generative model 245 based on the run of one or more validation datasets.
[0079] Once the generative model 245 is trained and / or validated, it may be stored in the knowledge base 240 and deployed to a local server 215 for runtime analysis of data communicated by devices 210 (i.e., data generated and / or collected and transmitted by devices 210 that were not used to train the generative model—for simplicity of explanation, this data is referred to herein as subsequent data). Deployment of the generative model 245 on the local server may be implemented via a deployment schedule, rule or policy-based, or some other similar deployment means. Once deployed, the generative model 245 is stored and implemented via a runtime AI engine 250 adapted to orchestrate a service or software application that includes the use of the generative model 245. These services or software applications may be provided to users of client computing devices as web-based or cloud services, for example, under an AaaS or SaaS model.
[0080] As part of a service or software application, the generative model 245 is used by the runtime AI engine 250 to analyze subsequent data and generate inferences or predictions from the analysis of the subsequent data. Inference refers to learning about the data generation process using the generative model 245 (e.g., inferring how a subject's disease state is influenced by the size and location of a tumor). Prediction, on the other hand, refers to predicting the outcome of a new data point using the generative model 245 (e.g., predicting a partial oxygen level from current vital signs and ventilator settings). The inference or prediction is then communicated to device 210 for use in one or more actions performed by device 210. One or more actions can be as simple as storing or displaying the inference or prediction on device 210 for use in controlling one or more parameters of device 210 (e.g., controlling ventilator parameters using predicted partial oxygen levels) or for more complex actions such as diagnosis and recommendation or administration of treatment based on the diagnosis.
[0081] IV. Technologies for deploying services on a digital health platform Figures 3–5 illustrate processes and operations for privacy-preserving computing to protect subject privacy while using subject data for secondary purposes such as training and deploying artificial intelligence tools. Individual embodiments may be described as processes shown as flowcharts, flow diagrams, data flow diagrams, structural diagrams, or block diagrams. While flowcharts describe operations as sequential processes, many operations may be performed in parallel or simultaneously. Furthermore, the order of operations may be reordered. A process terminates when its operations are complete, but may have additional steps not shown in the diagrams. A process may correspond to a method, function, procedure, subroutine, subprogram, etc. If a process corresponds to a function, its termination may correspond to the function's return to the calling function or main function.
[0082] The processes and / or operations shown in Figures 3 to 5 may be implemented by software (e.g., code, instructions, programs), hardware, or a combination thereof, executed by one or more processing units (e.g., processor cores). The software may be stored in memory (e.g., on a memory device, on a non-temporary computer-readable storage medium). The specific sequence of processing steps in Figures 3 to 5 is not intended to be limiting. Other sequences of steps may be performed according to alternative embodiments. For example, in alternative embodiments, the steps outlined above may be performed in a different order. Furthermore, the individual steps illustrated in Figures 3 to 5 may include multiple substeps, which may be performed in various sequences as appropriate for the individual steps. Furthermore, additional steps may be added or removed depending on the particular application. Those skilled in the art will recognize many variations, modifications, and substitutions.
[0083] Figure 3 shows flowchart 300 for privacy-preserving computing in a digital health platform. The process shown in flowchart 300 is implemented by the architecture, systems, and technologies shown in Figures 1 and 2.
[0084] In step 305, data (e.g., healthcare data) is collected by a device (e.g., device 210 as described in relation to Figure 2). In some cases, the data includes individually identifiable health information. The device may be operated actively or passively by a user, in which case it generates and / or collects data. In some cases, the data is deidentified by the device. In some cases, the data is collected by one or more devices (e.g., multiple devices operated by the same user but different users).
[0085] In step 310, the data is sent by the device to a local server (e.g., local server 215 as described in relation to Figure 2). The local server is part of a distributed computing environment (e.g., a cloud computing environment), such as a local cloud server. The local server is maintained within the data source region. The region can be a geographical region (e.g., a country or state), and the local server may be physically located in the same region as the user or device that generates and collects the data. In some cases, one or more devices are located within the same region and send their collected data to the same local server (regional-majority: 1). In other examples, one or more devices are distributed across multiple regions and send their collected data to local servers assigned to each region (regional-majority: 1, and globally-majority: many). Each region collectively shares a set of data rules regarding the use and storage of subject data, and the local server processing is adapted to use and store the data according to the set of data rules.
[0086] In step 315, the local server stores the data (raw data from the device) in one or more local data stores. The local server stores the data according to a set of data rules. For example, if a set of data rules states that individually identifiable health information must be deidentified during storage, the local server performs a deidentification operation on the data before storage. Additionally or alternatively, if a set of data rules states that individually identifiable health information must be anonymized before storage, the local server performs an anonymization operation on the data before storage. The data may be stored for subsequent processing or transmission as part of one or more operations or services provided by the local server. For example, the data may be stored in a data store before being input into one or more algorithms or models adapted to provide at least a part of the services provided by the local server.
[0087] In step 320, the local server processes the data and generates processed data (e.g., analytical data or anonymized data). The local server processes the data according to a set of data restrictions. For example, if a set of data restrictions states that individually identifiable health information must be deidentified before processing or transmission, the local server performs a deidentification operation on the data before processing or transmission. Additionally or alternatively, if a set of data restrictions states that individually identifiable health information must be anonymized before processing or transmission, the local server performs an anonymization operation on the data before processing or transmission. The data may be processed by the local server as part of one or more operations or services provided by the local server. For example, processing may include (i) inputting the data into one or more algorithms or models adapted to generate analytical data provided as at least part of the services provided by the local server, and / or (ii) sending the deidentified or anonymized data to one or more remote servers for storage and / or downstream processing (e.g., training one or more algorithms or models).
[0088] In step 325, the local server stores the processed data (e.g., analytical data or anonymized data) in one or more local data stores. The local server stores the processed data according to a set of data rules. The data may be stored for subsequent processing or transmission as part of one or more operations or services provided by the local server. For example, the data may be stored in a data store before being sent to one or more remote servers for storage and / or downstream processing.
[0089] In step 330, the processed data is sent by the local server to a remote server (e.g., remote server 220 as described in Figure 2). The remote server is part of a distributed computing environment (e.g., a cloud computing environment), such as a remote cloud server. The remote server is maintained in the same or a different region as the data source. The region can be a geographical region (e.g., a country or state), and the remote server may be physically located in the same or a different region as the user or device that generates and collects the data. In some cases, one or more local servers are located in the same region and send their processed data to a single remote server located in the same or a different region as one of the one or more local servers (territorial-majority: 1). In other examples, one or more local servers are distributed across multiple regions and send their processed data to one or more remote servers assigned to one or more regions (territorial-majority: 1, globally-majority: many). Each region collectively shares a set of data rules regarding the use and storage of subject data, and the remote server processing is adapted to use and store the data according to the set of data rules.
[0090] In step 335, the remote server identifies the version of the processed data and associates the processed data with a versioned dataset based on the identified version. If a versioned dataset is not available for association, a new versioned dataset is created for the identified version, and the processed data is associated with the new versioned dataset. The association of the processed data with a versioned dataset imposes version control on the processed data. As part of version control, the remote server determines the expiration date of the versioned dataset and stores the versioned dataset with an expiration date. The expiration date may be determined based on the creation date of the versioned dataset, the date of receipt of the processed data, the expiration date of the informed consent document associated with the processed data, or a combination thereof. The versioned dataset is stored in one or more data stores, either local or remote to the remote server.
[0091] In step 340, the remote server trains one or more algorithms and / or models using the versioned dataset (e.g., a training subset of the versioned dataset).
[0092] The model training process can involve performing iterative actions, selecting hyperparameters to input data into the model from one or more training datasets to find a set of model parameters (e.g., weights and / or biases) that minimizes one or more loss or error functions of the model. Hyperparameters are settings that can be tuned or optimized to control the behavior of the model. Most models explicitly define hyperparameters that control algorithms and / or different aspects of the model, such as memory or execution cost. However, additional hyperparameters may be defined to fit the model to a particular scenario. For example, hyperparameters can include the number of hidden units in the model, the learning rate of the model, the convolutional kernel width, or the number of kernels in the model. Each iteration of training can involve finding a set of model parameters (composed of a defined set of hyperparameters) of the model such that the value of one or more loss or error functions using a set of model parameters is less than the value of one or more loss or error functions using a different set of model parameters in the previous iteration. One or more loss or error functions may be constructed to measure the difference between the output inferred using the model (e.g., segmentation boundaries around one or more instances of an object of interest) and the ground truth (e.g., segmentation boundaries annotated on an image using labels).
[0093] Once a set of model parameters is identified, the model can be validated using a trained and versioned dataset (e.g., a test or validation subset of the versioned dataset). The validation process involves iterative operations inputting data from the versioned dataset into the model using validation techniques such as K-fold cross-validation, leave-one-out cross-validation, leave-one-group-out cross-validation, and nested cross-validation to tune the hyperparameters and ultimately find the optimal set of hyperparameters. Once the optimal set of hyperparameters is obtained, a reserved test set of data from the versioned dataset is input into the model to obtain an output (e.g., a segmentation boundary around one or more objects of interest), and the output is evaluated against the ground truth segmentation boundary by calculating performance metrics such as error, precision, precision, recall, and receiver operating characteristic curve (ROC) using correlation techniques such as Bland-Altman and Spearman's rank correlation coefficient.
[0094] In step 345, once the algorithm and / or model are trained and validated, the remote server identifies the version of the algorithm and / or model and associates the algorithm and / or model with a set of versioned algorithms and / or models based on the identified version. For example, the remote server may identify the version of the algorithm and / or model as version X. The remote server associates the algorithm and / or model with the set of algorithms and / or models of version X. If a set of versioned algorithms and / or models is not available for association, a new set of versioned algorithms and / or models is created for the identified version, and the algorithm and / or model is associated with the new set of versioned algorithms and / or models. The association between the new set of versioned algorithms and / or models and the set of versioned algorithms and / or models provides version control for the algorithm and / or model. As part of version control, the remote server determines the expiration date of the versioned set of algorithms and / or models and stores the versioned set of algorithms and / or models that have an expiration date. As an additional part of version control, a versioned set of algorithms and / or models is mapped to a versioned dataset used to train and validate the versioned set of algorithms and / or models. This mapping or association is then stored together with the versioned dataset and / or the versioned set of algorithms and / or models. Expiration dates may be determined based on the training date of the versioned set of algorithms and / or models, the validation date of the versioned set of algorithms and / or models, the expiration date of the associated versioned dataset, or a combination thereof.The versioned set of algorithms and / or models is stored in one or more local or remote data stores relative to the remote server.
[0095] In step 350, as an additional part of version control, the remote server removes or deletes existing versioned datasets and versioned sets of algorithms and / or models from one or more data stores based on their respective expiration dates. As understood, the version control implemented by the remote server may be adapted to remove or delete existing related versioned datasets and versioned sets of algorithms and / or models simultaneously based on their respective expiration dates, so that no algorithms and / or models exist in one or more data stores without the related versioned datasets used to train and validate the algorithms and / or models.
[0096] In step 355, the trained and validated (referred to as generated) algorithms and / or models selected from the versioned set of algorithms and / or models are sent to one or more local servers for deployment.
[0097] In step 360, the local server stores the generation algorithm and / or model in one or more local data stores.
[0098] In step 365, as described in step 305, subsequent data (e.g., healthcare data) is collected by the device. Subsequent data is data not used to train or validate the generation algorithm and / or model, but it may be used later to generate inference or predictions (as described in step 375). Subsequent data may be generated and / or collected from the same or different users and devices as the data generated and / or collected in step 305. To be understood, subsequent data can be any data collected by devices used by the user, as described in step 305.
[0099] In step 370, as described in step 310, the subsequent data is sent by the device to the local server.
[0100] In step 375, the local server stores and processes the subsequent data (raw data from the device and processed data from the operation or processor) as described in steps 315-335. In some cases, processing may involve inputting the subsequent data into one or more generative algorithms and / or models. For example, as part of a service or software application provided by the local server, one or more generative algorithms and / or models are used to analyze the subsequent data and generate inferences or predictions from the analysis of the subsequent data.
[0101] In step 380, inferences or predictions from the subsequent data analysis are stored in one or more data stores.
[0102] In step 385, the inference or prediction is communicated to the device for use in one or more actions performed by the device. The device may be the same device that generated and / or collected the subsequent data, or a different device.
[0103] In step 390, the device performs one or more actions using inference or prediction. These actions may be as simple as using the inference or prediction on the device to control one or more parameters of the device (e.g., using predicted partial oxygen values to control ventilator parameters) or to store or display the inference or prediction on the device for more complex actions such as diagnosis and recommendation or administration of treatment based on the diagnosis. In some cases, the actions may trigger or cause the generation of feedback data regarding the inference or prediction. For example, positive predicted values of the inference or prediction may be generated as a result of using the inference or prediction in one or more actions.
[0104] In step 395, feedback is collected by the device, sent to a local server, and stored by the local server in one or more data stores. The feedback itself may be in the form of subject data. In addition to inference or prediction, feedback may include data about any responses or activities triggered based on one or more actions in step 390. For example, feedback may include actions to be taken by the subject or provider based on inference or prediction. In some cases, feedback may indicate that the subject will undergo a digital examination (e.g., a vision test), adjust their diet, take medication, or record their blood pressure. Feedback may additionally trigger a trigger or machine to automatically record a measurement (e.g., O2 saturation). Feedback may be used in downstream processing to determine the effectiveness of a generative algorithm and / or model, and / or to subsequently train the generative algorithm and / or model.
[0105] Figure 4 illustrates a process 400 for privacy-preserving computing performed by a local cloud server on subject data used to develop artificial intelligence tools. In step 405, subject data relating to a first subject is received from a first computing device associated with the first subject. The subject data may be health management data containing individually identifiable health information. The first computing device may be a clinical device sensor, a handheld portable device, or a combination thereof. The local cloud server may be physically located in the same geographical region as the subject. The same geographical region may be the same country. In some cases, the same geographical region collectively shares a set of data regulations regarding the use and storage of individually identifiable health information.
[0106] In step 410, deidentification, anonymization, or both are performed on the subject data to generate processed subject data. Deidentification, anonymization, or both may be performed on individually identifiable health information of the subject data based on a set of data regulations (e.g., HIPAA regulations). In some cases, before performing deidentification, anonymization, or both on the subject data, the subject data is stored in a raw data store accessible to a local cloud server.
[0107] In step 415, the processed subject data is stored in a processed data store accessible from the local cloud server. The processed data store is not accessible from the remote cloud server.
[0108] In step 420, a batch of data containing processed subject data is sent to the remote cloud server. The batch data transmission may occur periodically or probabilistically so that the batch of data includes data from multiple other subjects captured since previous transmissions of data to the remote cloud server. In some cases, the transmission occurs in response to the local cloud server not receiving a request to delete the processed subject data before the processed subject data is transmitted. In other cases, a request to delete the subject data is received from the remote cloud server, and in response to the receipt of the request, the subject data is deleted from the raw data store. However, the processed subject data does not necessarily have to be deleted from the processed data store. In other examples, a request to delete the processed subject data is received from the remote cloud server, and in response to the receipt of the request, the processed subject data is deleted from the processed data store.
[0109] In step 425, the generative model is received from a remote cloud server. The generative model includes parameters partially derived from the processed subject data. The parameters may be derived based on minimizing one or more loss functions.
[0110] In step 430, subsequent data relating to a second subject is received from a second computing device associated with the second subject. The subsequent data may be subsequent healthcare data containing individually identifiable health information. The first computing device is the same or a different device as the second computing device. The second computing device may be a clinical device sensor, a handheld portable device, or a combination thereof. The first user is the same or a different user as the second user. In some cases, deidentification, anonymization, or both are performed on the subsequent data to generate processed subsequent data, which is stored in a processed data store accessible from a local cloud server, and batches of data containing the processed subsequent data are sent to a remote cloud server.
[0111] In step 435, subsequent data is input into the generative model, which analyzes the subsequent data and generates inferences or predictions from the analysis of the subsequent data. The inferences or predictions may be generated with respect to diagnosis, prognosis, treatment or therapy, identification of treatment or therapy protocols, detection or determination of disease status, identification or detection of biomarkers, reduction of non-compliance with treatment or therapy, reduction of operational costs, image analysis, marketing of treatment or therapy, automation of administrative tasks, assistance with medical procedures, or any combination thereof. Batches of subsequent data containing inferences or predictions may be sent to a remote cloud server. In some cases, deidentification operations, anonymization operations, or both are performed on the inferences or predictions to generate processed output data, which is stored in a processed data store accessible on the local cloud server, and batches of data containing the processed output data are sent to the remote cloud server.
[0112] In step 440, the inference or prediction is transmitted to the second computing device, the third computing device, or both for use in one or more actions performed by the second computing device, the third computing device, or a combination thereof. One or more actions may include communicating or displaying the inference or prediction, analyzing the inference or prediction, providing a procedure or treatment, initiating a procedure or treatment protocol, measuring biomarkers, providing notifications or reminders for a procedure or treatment, acquiring healthcare data, reporting a diagnosis or prognosis, analyzing images, providing marketing related to a procedure or treatment, controlling a medical device or medical procedure, or any combination thereof. In some cases, response data is received by a local cloud server from the second computing device, the third computing device, or both. Response data is generated in response to the receipt of the inference or prediction, or using the inference or prediction in one or more actions. De-identification operations, anonymization operations, or both are performed on the response data to generate processed response data. The processed response data is stored in a processed data store, and batches of data containing the processed response data are sent to a remote cloud server. In some cases, a new generative model is received from a remote cloud server, containing parameters partially derived from the processed response data, and the generative model is replaced with the new one. This replacement may include deleting the generative model from the local cloud server.
[0113] Figure 5 illustrates process 500 for privacy-preserving computing performed by a remote cloud server on subject data used to develop artificial intelligence tools. In step 505, processed subject data associated with multiple different subjects is received from a local cloud server, and the processed subject data is deidentified, anonymized, or both. The processed subject data may be health management data containing individually identifiable health information. The local cloud server may be physically located in the same geographical region as the subjects. The remote cloud server may be physically located in the same or a different geographical region as the local cloud server. The same geographical region may be the same country. In some cases, the same geographical region collectively shares a set of data regulations regarding the use and storage of individually identifiable health information. Individually identifiable health information in the processed subject data may be deidentified, anonymized, or both, based on the set of data regulations.
[0114] In step 510, the processed subject dataset is associated with a versioned dataset. In some cases, the association includes identifying a version of the processed subject dataset and associating the processed subject dataset with the versioned dataset based on the identified version of the processed subject dataset. If a versioned dataset is not available for association, a new versioned dataset is created for the identified version of the processed subject dataset, and the processed subject dataset is associated with the new versioned dataset.
[0115] In step 515, the expiration date of the versioned dataset is determined. The expiration date is determined based on the creation date of the versioned dataset, the date of receipt of the processed subject data, the expiration date of the informed consent document associated with the processed subject data, or any combination thereof.
[0116] In step 520, the versioned dataset is stored in a version data store accessible on a remote cloud server, associated with its expiration date. The version data store is not accessible on the local cloud server.
[0117] In step 525, the generative model is trained using the versioned dataset. The generative model includes parameters derived during training based on minimizing one or more loss functions.
[0118] In step 530, the association between the versioned dataset and the generative model is stored in the versioned datastore. The generative model may also be stored in the versioned datastore. In some cases, the versioned dataset and generative model are deleted or removed from the versioned datastore in response to the expiration of their validity period.
[0119] In step 535, the generative model is sent to a local cloud server for use in analyzing subsequent data and generating inferences or predictions from the analysis of the subsequent data. The subsequent data may be subsequent healthcare data containing individually identifiable health information. The inferences or predictions may be generated with respect to diagnosis, prognosis, treatment or therapy, identification of treatment or therapy protocols, detection or determination of disease status, identification or detection of biomarkers, reduction of non-compliance with treatment or therapy, reduction of operational costs, image analysis, marketing of treatment or therapy, automation of administrative tasks, assistance with medical procedures, or any combination thereof.
[0120] In some cases, another generative model is trained using the versioned dataset and / or another versioned dataset, the associations between the versioned dataset and / or another versioned dataset and the other generative model are stored in the versioned data store, and the other generative model is sent to a local cloud server for use in analyzing subsequent data and generating inferences or predictions from the analysis of the subsequent data. The other generative model may also be sent to the local cloud server with instructions to replace the generative model with the other generative model.
[0121] Viewed. Further consideration. Some embodiments of this disclosure include a system comprising one or more data processors. In some embodiments, the system includes a non-temporary computer-readable storage medium comprising instructions, wherein when the instructions are executed on one or more data processors, the one or more data processors cause one or more data processors to execute some or all of one or more of the methods and / or some or all of one or more processes disclosed herein. Some embodiments of this disclosure include a computer program product tangibly embodied in a non-temporary machine-readable storage medium comprising instructions configured to cause one or more data processors to execute some or all of one or more of the methods and / or some or all of one or more processes disclosed herein.
[0122] The terms and expressions used are for illustrative purposes only, not limitation, and in using such terms and expressions, there is no intention to exclude any equivalent or part of any of the features shown and described, however it should be recognized that various modifications are possible within the scope of the invention described in the claims. Accordingly, although the invention described in the claims is specifically disclosed by embodiments and optional features, modifications and variations of the concepts disclosed herein may be relied upon by those skilled in the art, and it should be understood that such modifications and variations are considered to be within the scope of the invention as defined by the appended claims.
[0123] The subsequent description provides only preferred exemplary embodiments and is not intended to limit the scope, applicability, or configuration of the Disclosure. Rather, the subsequent description of preferred exemplary embodiments provides possible descriptions for implementing various embodiments for those skilled in the art. It will be understood that various modifications can be made to the function and arrangement of the elements without departing from the spirit and scope set forth in the appended claims.
[0124] Specific details are given in the following description to provide a complete understanding of the embodiments. However, it will be understood that embodiments can be carried out without these specific details. For example, circuits, systems, networks, processes, and other components may be shown as components in block diagram form so as not to obscure the embodiments with unnecessary detail. In other examples, well-known circuits, processes, algorithms, structures, and techniques may be shown without unnecessary detail to avoid obscuring the embodiments.
Claims
1. 1. A computer-implemented method executed by a local cloud server, comprising: receiving subject data relating to a first subject from a first computing device associated with the first subject; performing a de-identification operation, anonymization operation, or both, on the subject data to generate processed subject data; storing the processed subject data in a processed data store accessible to the local cloud server; transmitting a batch of data to a remote cloud server, the batch of data including the processed subject data; receiving a generative model from the remote cloud server, the generative model including parameters derived in part from the processed subject data; receiving subsequent data regarding the second subject from a second computing device associated with the second subject; inputting the subsequent data into the generative model to analyze the subsequent data and generate an inference or prediction from the analysis of the subsequent data; transmitting the inference or the prediction to the second computing device, the third computing device, or both, for use in one or more operations performed by the second computing device, a third computing device, or a combination thereof; 4. A computer-implemented method comprising:
2. The computer-implemented method of claim 1 , wherein the subject data is healthcare data including individually identifiable health information, and the subsequent data is subsequent healthcare data including individually identifiable health information.
3. The local cloud server is physically located in the same geographic region as the subject; the same geographic regions collectively share a set of data regulations regarding the use and storage of the individually identifiable health information; 3. The computer-implemented method of claim 2, wherein the de-identification operation, the anonymization operation, or both, are performed on the individually identifiable health information of the subject data based on the set of data constraints.
4. The computer-implemented method of claim 1 , wherein the first computing device is the same or a different device as the second computing device.
5. The computer-implemented method of claim 4 , wherein the first computing device is a clinical device sensor, a handheld portable device, or a combination thereof.
6. The computer-implemented method of claim 4 or 5, wherein the second computing device is a clinical device sensor, a handheld portable device, or a combination thereof.
7. The computer-implemented method of claim 4 , wherein the first subject is the same or a different subject as the second subject.
8. The computer-implemented method of claim 1 , wherein the processed data store is inaccessible to the remote cloud server.
9. 9. The computer-implemented method of claim 1, wherein transmitting the processed subject data to the remote cloud server as part of the batch of data occurs in response to the local cloud server not receiving a request to delete the processed subject data prior to transmitting the processed subject data.
10. storing the subject data in a raw data store accessible to the local cloud server prior to performing the de-identification operation, the anonymization operation, or both on the subject data; receiving a request from the remote cloud server to delete the subject data; in response to receiving the request to delete the subject data, deleting the subject data from the raw data store; The computer-implemented method of claim 1 , further comprising:
11. The computer-implemented method of claim 10 , wherein the processed subject data is not deleted from the processed data store.
12. 12. The computer-implemented method of claim 1, wherein transmitting the processed subject data as part of the batch of data is performed periodically or probabilistically such that the batch of data includes data from a plurality of other subjects captured since a previous transmission of data to the remote cloud server.
13. 13. The computer-implemented method of any one of claims 1 to 12, wherein the inference or prediction is generated with respect to diagnosis, prognosis, treatment or therapy, identification of a treatment or therapy protocol, detection or determination of a disease state, identification or detection of a biomarker, reduction of non-adherence to a treatment or therapy, reduction of operational costs, image analysis, marketing of a treatment or therapy, automation of administrative tasks, assistance in medical procedures, or any combination thereof.
14. 14. The computer-implemented method of claim 1, wherein the one or more operations include communicating or displaying the inference or prediction, analyzing the inference or prediction, providing a treatment or therapy, initiating a treatment or treatment protocol, measuring a biomarker, providing notifications or reminders for a treatment or therapy, obtaining healthcare data, reporting a diagnosis or prognosis, analyzing images, providing marketing regarding a treatment or therapy, controlling a medical device or procedure, or any combination thereof.
15. 15. The computer-implemented method of claim 1, further comprising: transmitting a subsequent batch of data to the remote cloud server, the subsequent batch of data including the inference or the prediction.
16. performing the de-identification operation, the anonymization operation, or both, on the subsequent data to generate processed subsequent data; storing the processed subsequent data in the processed data store accessible to the local cloud server; transmitting a batch of data to the remote cloud server, the batch of data including the processed subsequent data; The computer-implemented method of claim 1 , further comprising:
17. performing the de-identification operation, the anonymization operation, or both, on the inferences or predictions to generate processed output data; storing the processed output data in the processed data store accessible to the local cloud server; transmitting a batch of data to the remote cloud server, the batch of data including the processed output data; The computer-implemented method of claim 1 , further comprising: receiving, at the local cloud server, from the second computing device, the third computing device, or both, response data generated in response to receiving the inference or prediction or using the inference or prediction in the one or more operations; performing the de-identification operation, the anonymization operation, or both, on the response data to generate processed response data; storing the processed response data in the processed data store; sending a batch of data to the remote cloud server, the batch of data including the processed response data; The computer-implemented method of claim 1 , further comprising:
19. receiving a new generative model from the remote cloud server, the new generative model including parameters derived in part from the processed response data; replacing the generative model with the new generative model, the replacing comprising deleting the generative model from the local cloud server; 20. The computer-implemented method of claim 18, further comprising:
20. 1. A system comprising: one or more data processors of a local cloud server; A non-transitory computer-readable storage medium containing instructions that, when executed on the one or more data processors, cause the one or more data processors to: receiving subject data relating to a first subject from a first computing device associated with the first subject; performing a de-identification operation, anonymization operation, or both, on the subject data to generate processed subject data; storing the processed subject data in a processed data store accessible to the local cloud server; transmitting a batch of data to a remote cloud server, the batch of data including the processed subject data; receiving a generative model from the remote cloud server, the generative model including parameters derived in part from the processed subject data; receiving subsequent data regarding the second subject from a second computing device associated with the second subject; inputting the subsequent data into the generative model to analyze the subsequent data and generate an inference or prediction from the analysis of the subsequent data; transmitting the inference or the prediction to the second computing device, the third computing device, or both, for use in one or more operations performed by the second computing device, a third computing device, or a combination thereof; A non-transitory computer-readable storage medium for causing the computer to perform operations including: A system comprising:
21. 21. The system of claim 20, wherein the subject data is healthcare data including individually identifiable health information, and the subsequent data is subsequent healthcare data including individually identifiable health information.
22. The local cloud server, wherein the local cloud server is physically located in the same geographic region as the subject; the same geographic regions collectively share a set of data regulations regarding the use and storage of the individually identifiable health information; 22. The system of claim 21, wherein the de-identification operation, the anonymization operation, or both, are performed on the individually identifiable health information of the subject data based on the set of data constraints.
23. 23. The system of claim 20, wherein the first computing device is the same or a different device as the second computing device.
24. 24. The system of claim 23, wherein the first computing device is a clinical device sensor, a handheld portable device, or a combination thereof.
25. 25. The system of claim 23 or 24, wherein the second computing device is a clinical device sensor, a handheld portable device, or a combination thereof.
26. 26. The system of any one of claims 23 to 25, wherein the first subject is the same or a different subject as the second subject.
27. 27. The system of claim 20, wherein the processed data store is inaccessible to the remote cloud server.
28. 28. The system of any one of claims 20 to 27, wherein transmitting the processed subject data to the remote cloud server as part of the batch of data occurs in response to the local cloud server not receiving a request to delete the processed subject data prior to transmitting the processed subject data.
29. The operation, storing the subject data in a raw data store accessible to the local cloud server prior to performing the de-identification operation, the anonymization operation, or both on the subject data; receiving a request from the remote cloud server to delete the subject data; in response to receiving the request to delete the subject data, deleting the subject data from the raw data store; 29. The system of any one of claims 20 to 28, further comprising:
30. 30. The system of claim 29, wherein the processed subject data is not deleted from the processed data store.
31. 31. The system of any one of claims 20 to 30, wherein transmitting the processed subject data as part of the batch of data is performed periodically or probabilistically such that the batch of data includes data from a plurality of other subjects captured since a previous transmission of data to the remote cloud server.
32. 32. The system of any one of claims 20 to 31, wherein the inference or prediction is generated with respect to diagnosis, prognosis, treatment or therapy, identification of a treatment or therapy protocol, detection or determination of a disease state, identification or detection of a biomarker, reduction of non-adherence to a treatment or therapy, reduction of operational costs, image analysis, marketing of a treatment or therapy, automation of administrative tasks, assistance in medical procedures, or any combination thereof.
33. 33. The system of any one of claims 20 to 32, wherein the one or more operations include communicating or displaying the inference or prediction, analyzing the inference or prediction, providing a treatment or therapy, initiating a treatment or treatment protocol, measuring a biomarker, providing notifications or reminders for a treatment or therapy, obtaining healthcare data, reporting a diagnosis or prognosis, analyzing images, providing marketing regarding a treatment or therapy, controlling a medical device or medical procedure, or any combination thereof.
34. 34. The system of claim 20, wherein the operations further comprise: transmitting a subsequent batch of data to the remote cloud server, the subsequent batch of data including the inference or the prediction.
35. The operation, performing the de-identification operation, the anonymization operation, or both, on the subsequent data to generate processed subsequent data; storing the processed subsequent data in the processed data store accessible to the local cloud server; transmitting a batch of data to the remote cloud server, the batch of data including the processed subsequent data; 34. The system of any one of claims 20 to 33, further comprising:
36. The operation, performing the de-identification operation, the anonymization operation, or both, on the inferences or predictions to generate processed output data; storing the processed output data in the processed data store accessible to the local cloud server; transmitting a batch of data to the remote cloud server, the batch of data including the processed output data; 34. The system of any one of claims 20 to 33, further comprising:
37. The operation, receiving, at the local cloud server, from the second computing device, the third computing device, or both, response data generated in response to receiving the inferences or predictions or using the inferences or predictions in the one or more operations; performing the de-identification operation, the anonymization operation, or both, on the response data to generate processed response data; storing the processed response data in the processed data store; sending a batch of data to the remote cloud server, the batch of data including the processed response data; 37. The system of any one of claims 20 to 36, further comprising:
38. The operation, receiving a new generative model from the remote cloud server, the new generative model including parameters derived in part from the processed response data; replacing the generative model with the new generative model, the replacing comprising deleting the generative model from the local cloud server; 38. The system of claim 37, further comprising:
39. 1. A computer program product for causing one or more data processors of a local cloud server to perform operations, the operations comprising: receiving subject data relating to a first subject from a first computing device associated with the first subject; performing a de-identification operation, anonymization operation, or both, on the subject data to generate processed subject data; storing the processed subject data in a processed data store accessible to the local cloud server; transmitting a batch of data to a remote cloud server, the batch of data including the processed subject data; receiving a generative model from the remote cloud server, the generative model including parameters derived in part from the processed subject data; receiving subsequent data regarding the second subject from a second computing device associated with the second subject; inputting the subsequent data into the generative model to analyze the subsequent data and generate an inference or prediction from the analysis of the subsequent data; transmitting the inference or the prediction to the second computing device, the third computing device, or both, for use in one or more operations performed by the second computing device, a third computing device, or a combination thereof; A computer program comprising:
40. 40. The computer program product of claim 39, wherein the subject data is healthcare data including individually identifiable health information, and the subsequent data is subsequent healthcare data including individually identifiable health information.
41. The local cloud server, wherein the local cloud server is physically located in the same geographic region as the subject; the same geographic regions collectively share a set of data regulations regarding the use and storage of the individually identifiable health information; 41. The computer program product of claim 40, wherein the de-identification operation, the anonymization operation, or both, are performed on the individually identifiable health information of the subject data based on the set of data constraints.
42. 42. The computer program product of any one of claims 39 to 41, wherein the first computing device is the same or a different device as the second computing device.
43. 43. The computer program product of claim 42, wherein the first computing device is a clinical device sensor, a handheld portable device, or a combination thereof.
44. 44. The computer program product of claim 42 or 43, wherein the second computing device is a clinical device sensor, a handheld portable device, or a combination thereof.
45. 45. The computer program of any one of claims 42 to 44, wherein the first subject is the same or a different subject as the second subject.
46. 46. The computer program product of claim 39, wherein the processed data store is inaccessible to the remote cloud server.
47. 47. The computer program product of any one of claims 39 to 46, wherein transmitting the processed subject data to the remote cloud server as part of the batch of data is performed in response to the local cloud server not receiving a request to delete the processed subject data prior to transmitting the processed subject data.
48. The operation, storing the subject data in a raw data store accessible to the local cloud server prior to performing the de-identification operation, the anonymization operation, or both on the subject data; receiving a request from the remote cloud server to delete the subject data; in response to receiving the request to delete the subject data, deleting the subject data from the raw data store; 48. The computer program of any one of claims 39 to 47, further comprising:
49. 49. The computer program of claim 48, wherein the processed subject data is not deleted from the processed data store.
50. 50. The computer program product of any one of claims 39 to 49, wherein transmitting the processed subject data as part of the batch of data is performed periodically or probabilistically such that the batch of data includes data from a plurality of other subjects captured since a previous transmission of data to the remote cloud server.
51. 51. The computer program of any one of claims 39 to 50, wherein the inference or the prediction is generated with respect to diagnosis, prognosis, treatment or therapy, identification of a treatment or therapy protocol, detection or determination of a disease state, identification or detection of a biomarker, reduction of non-adherence to a treatment or therapy, reduction of operational costs, image analysis, marketing of a treatment or therapy, automation of administrative tasks, assistance in medical procedures, or any combination thereof.
52. 52. The computer program of any one of claims 39 to 51, wherein the one or more operations comprise communicating or displaying the inference or prediction, analyzing the inference or prediction, providing a treatment or therapy, initiating a treatment or treatment protocol, measuring a biomarker, providing notifications or reminders for a treatment or therapy, obtaining healthcare data, reporting a diagnosis or prognosis, analyzing images, providing marketing regarding a treatment or therapy, controlling a medical device or medical procedure, or any combination thereof.
53. 53. The computer program product of claim 39, wherein the operations further comprise: transmitting a subsequent batch of data to the remote cloud server, the subsequent batch of data including the inference or the prediction.
54. The operation, performing the de-identification operation, the anonymization operation, or both, on the subsequent data to generate processed subsequent data; storing the processed subsequent data in the processed data store accessible to the local cloud server; transmitting a batch of data to the remote cloud server, the batch of data including the processed subsequent data; 53. The computer program of any one of claims 39 to 52, further comprising:
55. The operation, performing the de-identification operation, the anonymization operation, or both, on the inferences or predictions to generate processed output data; storing the processed output data in the processed data store accessible to the local cloud server; transmitting a batch of data to the remote cloud server, the batch of data including the processed output data; 53. The computer program of any one of claims 39 to 52, further comprising:
56. The operation, receiving, at the local cloud server, from the second computing device, the third computing device, or both, response data generated in response to receiving the inferences or predictions or using the inferences or predictions in the one or more operations; performing the de-identification operation, the anonymization operation, or both, on the response data to generate processed response data; storing the processed response data in the processed data store; sending a batch of data to the remote cloud server, the batch of data including the processed response data; 56. The computer program of any one of claims 39 to 55, further comprising:
57. The operation, receiving a new generative model from the remote cloud server, the new generative model including parameters derived in part from the processed response data; replacing the generative model with the new generative model, the replacing comprising deleting the generative model from the local cloud server; 57. The computer program of claim 56, further comprising:
58. 1. A computer-implemented method executed by a remote cloud server, comprising: receiving processed subject data associated with a plurality of different subjects from the local cloud server, the processed subject data being de-identified, anonymized, or both; Associating the processed subject dataset with the versioned dataset; determining an expiration date for the versioned data set; storing the versioned dataset in a version data store accessible to the remote cloud server, wherein the versioned dataset is stored in association with an expiration date; training a generative model using the versioned dataset; storing an association between the versioned dataset and the generative model trained with the versioned dataset in a versioned data store; transmitting the generative model to the local cloud server for use in analyzing subsequent data and generating inferences or predictions from the analysis of the subsequent data; 4. A computer-implemented method comprising:
59. 60. The computer-implemented method of claim 58, wherein the processed subject data is healthcare data including individually identifiable health information, and the subsequent data is subsequent healthcare data including individually identifiable health information.
60. 60. The computer-implemented method of claim 58 or 59, wherein the version data store is inaccessible to the local cloud server.
61. The local cloud server, wherein the local cloud server is physically located in the same geographic region as the subject; the same geographic regions collectively share a set of data regulations regarding the use and storage of the individually identifiable health information; 60. The computer-implemented method of claim 59, wherein the individually identifiable health information of the processed subject data has been de-identified, anonymized, or both based on the set of data constraints.
62. 62. The computer-implemented method of any one of claims 58 to 61, wherein the expiration date is determined based on a creation date of the versioned dataset, a receipt date of the processed subject data, an expiration date of an informed consent form associated with the processed subject dataset, or any combination thereof.
63. storing the Generative Model in the versioned data store accessible to the remote cloud server; deleting or removing the versioned dataset and the Generative Model from the versioned data store in response to the expiration date passing; 63. The computer-implemented method of any one of claims 58 to 62, further comprising:
64. training another generative model using the versioned dataset and / or another versioned dataset; and storing, in the versioned data store, an association between the versioned dataset and / or the other versioned dataset and the other generative model trained with the versioned dataset and / or the other versioned dataset; transmitting the other generative model to the local cloud server for use in analyzing the subsequent data and generating the inference or prediction from the analysis of the subsequent data, the other generative model being transmitted along with instructions to the local cloud server to replace the generative model with the other generative model; 64. The computer-implemented method of any one of claims 58 to 63, further comprising:
65. 65. The computer-implemented method of any one of claims 58 to 64, wherein the inference or prediction is generated with respect to diagnosis, prognosis, treatment or therapy, identification of a treatment or therapy protocol, detection or determination of a disease state, identification or detection of a biomarker, reduction of non-adherence to a treatment or therapy, reduction of operational costs, image analysis, marketing of a treatment or therapy, automation of administrative tasks, assistance in medical procedures, or any combination thereof.
66. 66. The computer-implemented method of any one of claims 58 to 65, wherein the generative model comprises parameters derived during the training based on minimizing one or more loss functions.
67. 1. A system comprising: one or more data processors in a remote cloud server; A non-transitory computer-readable storage medium containing instructions that, when executed on the one or more data processors, cause the one or more data processors to: receiving processed subject data associated with a plurality of different subjects from the local cloud server, the processed subject data being de-identified, anonymized, or both; Associating the processed subject dataset with the versioned dataset; determining an expiration date for the versioned data set; storing the versioned dataset in a version data store accessible to the remote cloud server, wherein the versioned dataset is stored in association with an expiration date; training a generative model using the versioned dataset; storing an association between the versioned dataset and the generative model trained with the versioned dataset in a versioned data store; transmitting the generative model to the local cloud server for use in analyzing subsequent data and generating inferences or predictions from the analysis of the subsequent data; A non-transitory computer-readable storage medium for causing the computer to perform operations including: A system comprising:
68. 68. The system of claim 67, wherein the processed subject data is healthcare data including individually identifiable health information, and the subsequent data is subsequent healthcare data including individually identifiable health information.
69. 69. The system of claim 67 or 68, wherein the version data store is not accessible to the local cloud server.
70. The local cloud server, wherein the local cloud server is physically located in the same geographic region as the subject; the same geographic regions collectively share a set of data regulations regarding the use and storage of the individually identifiable health information; 69. The system of claim 68, wherein the individually identifiable health information of the processed subject data has been de-identified, anonymized, or both based on the set of data constraints.
71. 71. The system of any one of claims 67-70, wherein the expiration date is determined based on a creation date of the versioned dataset, a receipt date of the processed subject data, an expiration date of an informed consent document associated with the processed subject dataset, or any combination thereof.
72. The operation, storing the Generative Model in the versioned data store accessible to the remote cloud server; deleting or removing the versioned dataset and the Generative Model from the versioned data store in response to the expiration date passing; 72. The system of any one of claims 67 to 71, further comprising:
73. The operation, training another generative model using the versioned dataset and / or another versioned dataset; and storing, in the versioned data store, an association between the versioned dataset and / or the other versioned dataset and the other generative model trained with the versioned dataset and / or the other versioned dataset; transmitting the other generative model to the local cloud server for use in analyzing the subsequent data and generating the inference or prediction from the analysis of the subsequent data, the other generative model being transmitted along with instructions to the local cloud server to replace the generative model with the other generative model; 73. The system of any one of claims 67 to 72, further comprising:
74. The system of any one of claims 67 to 73, wherein the inference or prediction is generated with respect to diagnosis, prognosis, treatment or therapy, identification of a treatment or therapy protocol, detection or determination of a disease state, identification or detection of a biomarker, reduction of non-adherence to a treatment or therapy, reduction of operational costs, image analysis, marketing of a treatment or therapy, automation of administrative tasks, assistance in medical procedures, or any combination thereof.
75. 75. The system of any one of claims 67 to 74, wherein the generative model comprises parameters derived during the training based on minimizing one or more loss functions.
76. 1. A computer program product for causing one or more data processors of a remote cloud server to perform operations, the operations comprising: receiving processed subject data associated with a plurality of different subjects from the local cloud server, the processed subject data being de-identified, anonymized, or both; Associating the processed subject dataset with the versioned dataset; determining an expiration date for the versioned data set; storing the versioned dataset in a version data store accessible to the remote cloud server, wherein the versioned dataset is stored in association with an expiration date; training a generative model using the versioned dataset; storing an association between the versioned dataset and the generative model trained with the versioned dataset in a versioned data store; transmitting the generative model to the local cloud server for use in analyzing subsequent data and generating inferences or predictions from the analysis of the subsequent data; A computer program comprising:
77. 77. The computer program product of claim 76, wherein the processed subject data is healthcare data including individually identifiable health information, and the subsequent data is subsequent healthcare data including individually identifiable health information.
78. 78. The computer program product of claim 76 or 77, wherein the version data store is not accessible to the local cloud server.
79. The local cloud server is physically located in the same geographic region as the subject; the same geographic regions collectively share a set of data regulations regarding the use and storage of the individually identifiable health information; 78. The computer program of claim 77, wherein the individually identifiable health information of the processed subject data has been de-identified, anonymized, or both based on the set of data constraints.
80. 80. The computer program of any one of claims 76 to 79, wherein the expiration date is determined based on a creation date of the versioned dataset, a receipt date of the processed subject data, an expiration date of an informed consent document associated with the processed subject dataset, or any combination thereof.
81. The operation, storing the Generative Model in the versioned data store accessible to the remote cloud server; deleting or removing the versioned dataset and the Generative Model from the versioned data store in response to the expiration date passing; 81. The computer program of any one of claims 76 to 80, further comprising:
82. The operation, training another generative model using the versioned dataset and / or another versioned dataset; and storing, in the versioned data store, an association between the versioned dataset and / or the other versioned dataset and the other generative model trained with the versioned dataset and / or the other versioned dataset; transmitting the other generative model to the local cloud server for use in analyzing the subsequent data and generating the inference or prediction from the analysis of the subsequent data, the other generative model being transmitted along with instructions to the local cloud server to replace the generative model with the other generative model; 82. The computer program of any one of claims 76 to 81, further comprising:
83. 83. The computer program of any one of claims 76 to 82, wherein the inference or the prediction is generated with respect to diagnosis, prognosis, treatment or therapy, identification of a treatment or therapy protocol, detection or determination of a disease state, identification or detection of a biomarker, reduction of non-adherence to a treatment or therapy, reduction of operational costs, image analysis, marketing of a treatment or therapy, automation of administrative tasks, assistance in medical procedures, or any combination thereof.
84. 84. The computer program of any one of claims 76 to 83, wherein the generative model comprises parameters derived during the training based on minimizing one or more loss functions.
85. A non-transitory machine-readable storage medium storing a computer program according to any one of claims 39 to 57.
86. A non-transitory machine-readable storage medium storing a computer program according to any one of claims 76 to 84.
