Smart and Secure PACS (Picture Archiving and Communication System) on Clouds

By Synho Do, Ph.D., Director, Laboratory of Medical Imaging and Computation, Assistant Professor, Massachusetts General Hospital and Harvard Medical School

With the digitization of healthcare comes the discovery of newly developed areas of Artificial Intelligence (AI) and Machine Learning (ML). Converting film to digital pictures in Radiology is equivalent to developing a fully electric car. While the initial worries and concerns of digital radiography create uncertainty of the unknown, those worries seemingly disappear when looking at the current success and essential role in healthcare. An undersized initial investment has opened up a method that can have an immediate impact and create further advancement in the field.

Digitally well-organized data can be easily processed with an algorithm called CNN (Convolutional Neural Network). Using a GPU (Graphics Processing Unit) capable of high-speed operation through parallel processing, it is possible to develop an AI algorithm with a basic programming skillset and simple understanding of models that accurately predict complex outcomes. One prerequisite is data well organized and clearly labeled as accurately as possible. In order to create the most optimal processing, the quality of data curation is more necessary than the quantity of data available.

When the previously developed Cryptographic technologies are tested and implemented in the healthcare data flow, value generation through AI algorithm will become possible.

Before the arrival of modern AI, problems were solved through conventional solution methods like mathematics and engineering in a well-organized format. However, in the latest approach, given a case of organized data, it is possible to create an algorithm by discovering unknown weighting in the topology of CNN architecture. In general, a complex neural network composed of many layers has adequate performance. Though more data is needed to train this network, in this case, the quantity of the data plays a crucial role in estimating weights in the AI/ML model.

And so, if it is necessary to determine the rank of importance, the quality of data comes first, followed by the amount of data. The third essential aspect is high-performance computing power that is able to process not only complex, but also large amounts of data. Many startups, hospitals, and governments are now well aware of the importance of data and are developing data-oriented science platforms supported by expertise, copious data, massive computational power, as well as funding from outside of the hospital.

Cloud solutions such as Amazon, Google, Microsoft, and Nvidia are trying to provide a high capacity while also achieving a secure computing environment. The advantage of the cloud is its easy scalability. Users do not have to worry about hardware problems. There are also various models of the price according to its usage.

Currently, uploading hospital data to the cloud seems to hold unnecessary risk as the system is very unpredictable. From a hospital’s perspective, the risk of leaking sensitive patient information holds a more serious danger as not only would the act be unethical, but it would also cause the group to be legally responsible. The risk is simply too high. Nevertheless, a more convenient and secure method in which patients can safely store their own data can be created. 

In part, personal healthcare information (PHI) can be collected through mobile devices or IoT (Internet of Things), linked with healthcare apps in the cloud. Many AI-related business models are being proposed, and to implement them, there is currently a race to obtain data of high quality and sufficient volume. Ultimately, the source and ownership of data become a new issue to consider. Could it be possible to propose a method to compensate patients according to the contribution of fairly distributed data? If this method is transparent and safe, will people voluntarily trust and use it?

While the issues that currently exist in the hospital setting seem relatively new, the same issue has been around for a long time in cryptography with solutions found. In many cases, innovation solves our problems by sharing successful technology. Techniques that work well in one field can be easily translated to another and it is the translation of one to another that we call growth in technology.

Cryptography, which studies how to encrypt data to send and receive information safely, started from the mathematical basis of number theory and is used in many application fields in a practical sense.

• How do you authenticate with the other party without leaking any information? Zero-knowledge authentication.
• Is it possible to confirm that even a tiny amount of data does not change in a massive amount of data? Hashing
• Can I confirm to others that I am the real me? DID (Decentralized Identifiers)
• Can everyone confirm that I am the valid owner of this digital content? NFT (Non-fungible token)
• Can you trust the result of the two data calculated after encrypting the critical data without exposing it? FHE (Fully Homomorphic Encryption)

When the previously developed Cryptographic technologies are tested and implemented in the healthcare data flow, value generation through AI algorithm will become possible. Those who provide data will be rewarded according to the accuracy and rarity of their data. Experts who label data can also be rewarded according to accuracy and difficulty. In addition, data scientists and entrepreneurs who safely manage and process data can be rewarded according to the performance and benefit of AI solutions. While the end goal will be a solution to our flawed hospital data systems, significant changes will come from many small successes.