Accelerating AI and ML in Healthcare Using Blockchain

At HIMSS18 last week AI (Artificial Intelligence) / ML (Machine Learning), and Blockchain were the topics of greatest interest. Not only are these high priority, strategic technologies poised to transform healthcare, but there is actually a lot of synergy between them, with blockchain providing a foundation and catalyst to accelerate the evolution AI and ML in healthcare. I had the privilege of moderating a blockchain session, Blockchain Reset – Seeing Through the Hype and Starting Down the Path, with a panel of experts. This session was a collaborative initiative with the HIMSS Blockchain Working Group, and highly successful with a full house of attendance and 3 rows standing at the back. We had many great questions and lively dialog during the Q&A and could have used much more time. This session was video recorded and will be made available for on-demand viewing via the HIMSS conference website shortly, likely linked from the previous session description URL. Watch my LinkedIn posts for an update on this once available. I also had many opportunities at HIMSS18 to meet with healthcare organizations and industry experts working on a variety of different use cases for AI / ML and blockchain in healthcare. There is wide recognition that AI / ML and blockchain will impact and disrupt healthcare across multiple use cases, and leaders from across all segments of healthcare are working to understand how to leverage these technologies to improve patient care, and reduce costs. In this article I want to share some key insights with you from these activities.

More Data, Higher Quality for Better AI / ML

AI / ML do better with more data volume, and higher quality data to train models. If we only source data from a single healthcare organization silo then we will only achieve limited results. Many use cases may not be feasible because due to a lack of high quality data we cannot achieve a low enough error rate. If, however, we are able to source data for AI / ML model training from across a network of healthcare organizations using blockchain then we could use the union of all of the data available across the network, and select higher quality data to train models, reducing error rates, maximizing benefits, and enabling more use cases. Blockchain can be used to post metadata, pointers, and hashcodes about data that exists in various silos within various healthcare organizations participating in a blockchain network. Other healthcare organizations in this network can use this information on blockchain to search, discover, locate, and subsequently securely retrieve data from peers as required to build specific new AI / ML models. Blockchain and digital signatures can be used to track provenance, authenticity, and integrity of data used to train AI / ML models.

Building AI / ML Models Using Blockchain

In building AI / ML models it is particularly important to use high quality data and minimize risk of biased models that could lead to errors. Blockchain enables tracking of data sets and any additional parameters used to build and train AI / ML models. If the provenance of data is tracked on blockchain, then this information can be taken into consideration in filtering out the highest quality data suitable to train models and conduct inference. Once models are built it is important to protect their integrity and be able to verify their authenticity and this can also be done using blockchain. This is particularly important in healthcare AI / ML use cases where models could be used to provide input to clinicians that need to make critical patient care decisions based on the inference from the models. Lastly, AI / ML models can be shared using blockchain, and more rapidly evolved and matured through collaboration across a network of healthcare organizations vs each healthcare organization working independently. Such sharing and collaboration can also reduce redundant work and help reduce healthcare costs. AI / ML models could be trained with both data stored on the blockchain as well as data stored off the blockchain (and referenced by the blockchain) in conventional enterprise systems.

Tracking and Validating Inference Results with Blockchain

Similarly, AI / ML models can conduct inference on both data stored on the blockchain as well as data stored off the blockchain (and referenced by the blockchain) in conventional enterprise systems. If AI / ML inference tells us what we expect then it will be easy to trust it ... but if it only tells us what we expect then it adds no value. If it tells us something slightly different to what we expect we may trust it over time, with enough positive experience and trust. However, if it tells us something radically different we are unlikely to trust it in the near term. This is a dilemma because if a model tells us what we expect to hear then it is really adding no value. However, if it tells us something radically new it could be a major breakthrough, but do we trust it? Trust in AI / ML results is built up over time from positive experiences with the associated models, inferences they produce, and positive validations thereof. If each healthcare organization builds up this experience independently and redundantly then it will take longer, and cost more collectively. If blockchain is used to enable sharing of AI / ML model inference results and validations of those results across a network of healthcare organizations, then we can more rapidly build trust in the model, evolve it, reduce costs, and pave the way for maximizing the value of inference and insights learned from the model. If we ever need to audit an inference result, either as part of a routine validation or an investigation, then blockchain can provide a complete audit trail of how the model was built from data and collaboration across the network of healthcare organizations, i.e. what data was used to train it, the provenance, integrity, and authenticity of the data, and so forth.

AI / ML To Power Smart Contracts and DAOs

Smart contracts can automate the processing of transactions on blockchain. Cryptocurrencies and tokens can run on top of smart contracts on the blockchain and enable new marketplaces and incentives to catalyze the sharing of data, models, results, validations and collaboration on advancing AI / ML. DAOs (Decentralized Autonomous Organizations) can be built from cryptocurrencies and smart contracts running on blockchain. Both smart contracts and DAO's can make use of AI / ML as part of logic that is executed, triggered by the arrival of new matching input transactions appended to the blockchain, and resulting in new output transactions that are in turn appended to the same blockchain.

Building Engagement and Trust with Patients

Patients, or data subjects in privacy speak, are arguably the most important entity to engage and build trust with in AI / ML as they are the source of data. Blockchain enables new levels of patient engagement enabling them to review their records, amend as needed, consent (opt-in / opt-out) to their participation in studies, be rewarded for their participation e.g. with cryptocurrencies and tokens, audit access to their data, and also benefit more directly from the results of clinical studies in which they participate. This is key to success for the existing healthcare ecosystem where the vast majority of patient data is within enterprise silos, within healthcare organizations, and fragmented across many healthcare organizations over the lifetime of the patient. This is also even more important going forward where more and more of the patient data will come from "consumer health", i.e. from the patient directly using various wearables and medical devices, and numerous healthcare services the patient engages with directly.

Conclusion

In summary, blockchain can enable the discovery, location, and sharing of AI / ML data, models, results, and validations. It can provide incentives for sharing and enable a marketplace using cryptocurrencies and tokens. More data from across the network of healthcare organizations, and higher quality data with detailed provenance records on blockchain, can be used to build better models, achieve better inference results, and maximize benefits in improve patient care, and reducing healthcare costs. Blockchain can provide transparency in where data comes from, how models are built, validations of inference results, and catalyze collaboration around AI / ML to build trust, and evolve, mature, and improve AI / ML collectively much faster and at much lower cost collectively than if each healthcare organization does this independently, and redundantly.

Getting Started with Blockchain

If you would like to get started with prototyping blockchain see the Azure Blockchain Workbench for a powerful platform for rapid prototyping of your blockchain, and subsequent deployment onto Ethereum Enterprise blockchain running in the Microsoft Azure cloud. The Azure Blockchain Workbench will also be adding support for Hyperledger Fabric and R3 Corda blockchain platforms going forward. This platform enables you to accelerate your technical POCs and pilots, and focus on your use case, business value, and pilot rather than blockchain technologies and deployment complexities.

Getting Started with AI

Bootstrap your healthcare AI initiative with the Azure Security and Compliance Blueprint - HIPAA/HITRUST Health Data and AI. Rather than starting from scratch, you can accelerate your AI initiative by rapidly downloading, configuring, running, and customizing this AI blueprint for your use case. This blueprint also provides a wealth of information on how to protect the privacy, security, and achieve HIPAA and HITRUST compliance with your cloud based AI solution.

Collaboration

What synergies do you see with AI / ML and blockchain? Welcome any feedback and comments below. Microsoft is actively working in these areas of innovation. Message me on LinkedIn if you would like to connect, discuss, and explore synergies and opportunities for collaboration.

Related

1. A Data Centric View of Blockchain
2. Blockchain: 6 Key Ethical Considerations
3. Blockchain CyberSecurity – What You Need to Know to Avoid a Breach
4. Food is Medicine – Will the first large scale production use of blockchain in healthcare be food supply chain?
5. 8 Opportunities to Advance AI in Healthcare Using Blockchain
6. Blockchain as a Tool for Anti-Fraud
7. Healthcare Blockchain Privacy
8. Blockchain in Healthcare: The Potential and Limitations
9. Blockchain in Healthcare Webinar: Patient Privacy & Cybersecurity in DLT Architecture, Planning, & Adoption
10. BlockRx Asks the Experts: David Houlding, Intel Health & Life Sciences
11. Will Blockchains Deliver Healthcare Interoperability?
12. Blockchain, Cryptocurrencies, Smart Contracts, Artificial Intelligence, and Machine Learning in Healthcare
13. Healthcare Use Cases for Blockchain - 5 Key Factors for Success
14. Healthcare Blockchain: What Goes On Chain Stays on Chain
15. Healthcare Blockchain: Does Your Chain Have any Weak Links?
16. Will Your Healthcare Blockchain be Available When you Need It?