AI (Artificial Intelligence) and ML (Machine Learning) have shown incredible potential in healthcare across a wide range of use cases, from diagnostic imaging, to anti-fraud, resource and asset optimization, readmission prevention, behavioral analytics, medical risk analytics, claims analytics, and many more. In a series of blogs, we will share a more detailed look at the opportunities to advance AI in healthcare using blockchain for those that want a more technical deep dive. For this first post, we’ll be covering ways that using blockchain in healthcare can help build higher quality models, obtain better data, improve auditing, and protect the integrity of the models.
Why use blockchain in healthcare?
Whether you are using AI/ML to optimize your operations or to improve patient care, the success of using this technology in each of these use cases hinges on the quality of inference that can be done and achieving acceptably low error rates. These in turn depend on the quality of the models. Building higher quality models in almost all cases can be done better through collaboration across a consortium of healthcare organizations instead of any one organization going at it alone. Using blockchain can help to address this issue and unlock the power of AI for healthcare organizations.
Four Ways Blockchain Can Advance AI in Health
- More training data from across a consortium and improved ability to specialize
AI / ML are extremely data hungry. The more training data, the better the models, the better the inference and results. Almost all AI / ML efforts are limited by data available. In most cases data used to train models is being sourced from just a single organization. Blockchain can be used to publish metadata about data that exists across a consortium of healthcare organizations. This metadata can include pointers to the enterprise systems that store the data, and hashcodes that can be used to verify the integrity of data. Organizations participating in such a blockchain can discover available data, locate it, and subsequently request data of interest via a secure, direct peer-to-peer exchange. Metadata on the blockchain can include information used to determine data of a particular specialty, eg x-rays of tumors of a particular kind. Having more data from across a consortium, and the ability to query by data specialty based on metadata on the blockchain enables new levels of specialization of data sets, and specialized models trained with it.
- Higher quality data and models through tracking provenance
Biasing of models is a common problem with AI / ML. In healthcare, a biased model can skew results, or increase error rates in ways that can impact inference results and ultimately patient care. Metadata on blockchain can include provenance information that enables the highest quality data sets from across the consortium to be identified and only this data included in training models to help mitigate biasing.
- Improved quality management through auditing
Blockchains excel at protecting the integrity of data. This makes them particularly well suited to storing audit trails that require such integrity protection to mitigate risk of accidents, fraud, and other risks to data integrity. Blockchain can be used to record all audit information regarding the building, testing, and use of AI / ML in healthcare. This can include training data, models and versions through the adaptive learning process, results generated, validations of results, who did what, when, where, why, how, and so forth. In the event of an incident, for example a biased model is detected, one can go to the audit trail and see exactly what data went into the model, root out data causing the biasing, retrain models, and correct the issue.
- Protecting the integrity of AI / ML
As healthcare grows to depend on AI / ML, so does the need to protect the integrity of models and other associated assets since corruption of these assets, whether accidental or malicious can impact results, and in a worst case such as in diagnostic imaging could directly impact patient care. As mentioned earlier, blockchains excel at protecting data integrity, and for all practical purposes they are immutable. Blockchains can protect both data stored on the blocks of the chain, as well as data stored off-chain and referenced by metadata, pointers, and hashcodes as discussed previously. In the latter case, the integrity of any record stored off-chain can be checked at any point through checking its hashcode against the hashcode stored on the blockchain for the record. If they don’t match integrity compromise is detected, data discarded, and an alert can be issued to initiate remediation.
These are just a few of the opportunities available to help accelerate AI in healthcare using blockchain. What other opportunities do you see?
AI, ML, and blockchain in healthcare are fast evolving. The intersection of these technologies is very new, and even faster evolving. Many of these new concepts do not yet appear in books. I post regularly about new developments in healthcare, AI / ML, blockchain, cloud computing, security, privacy, and compliance on social media. If you are a healthcare organization looking to implement AI / ML or blockchain, or if you are helping healthcare organizations get started with Microsoft technologies for AI / ML and blockchain and would like to explore partnership, we’d love to hear from you. You can find me on LinkedIn and Twitter.
Finally, if you’re ready to get started implementing blockchain and/or AI, take a look at these resources:
- Accelerate your AI / ML in healthcare initiative using this AI in Healthcare Blueprint which includes executable code, test data, automated deployment, and documentation that enables you to rapidly establish a working reference point for your solution in your Microsoft Azure cloud.
- Rapid prototype your blockchain solution using Azure Blockchain Workbench, and deploy to the Microsoft Azure cloud on Ethereum, to enable you to focus more on your blockchain solution rather than development and deployment complexities.