A New Journey into Healthcare

The following is a guest article by Deborah Millin, Senior Director of Healthcare Business Development at NVIDIA.

The coined term “Smart Hospital” has been the most talked about concept in Health IT journals in recent years since the introduction of HIPAA, illustrating the plethora of disruptive technologies that is casting a wide net over healthcare organizations who are proudly displaying their benefits like a badge of innovative honor.

Smart Hospital initiatives are supported by bountiful vendors vying for this ‘space’.  From remote monitoring, IV pumps, sensors, robotics, central video management, medical instruments, networking, RFID; the list goes on, but if there is one thing these “smart devices” have in common is that they produce data; and lots of it.

While the array of devices and technologies are certainly disruptive in their own right, the healthcare holy grail lies in the world of gaining actionable insight that allows clinicians the ability to discover, analyze and accelerate medicine in an unprecedented era since the advent of the internet. The presentation of data from all kinds of different sources can be potentially translated to more insightful outcomes, reducing cost and risk.

As we transcend to a world of value-based care, these factors contribute to the consideration of an ever-evolving adoption of Smart Hospital Strategies.  Valuable investments toward digitizing and extracting data for healthcare have been compared to mining for gold. But in the words of Ralph Waldo Emerson, “The desire of gold is not for gold, it is for the means of freedom and benefit.”

Analytics and Artificial Intelligence supported by Machine Learning(ML) and Deep Learning(DL) are gold nuggets whose benefits are seen through insightful and meaningful outcomes that aim to improve the quality and longevity of human life in the most efficient, safe and cost effective ways we have ever seen.  Newly developed concepts are often perceived as risk but the cost of resistance to adopting and exploiting those very technologies will end up leaving healthcare organizations behind, losing competitive advantage to those opting to modernize their digitization strategy, accomplishing what used to take weeks, often months, to just hours or minutes.

While we are bound by regulations, cost, skillsets and time, these very elements used as adoption barriers are now being hurdled by Smart Hospital strategy.

What is a Smart Hospital?

Over the years there has been an evolving field creating a checkerboard of adopted technologies offering their own unique value. We are now at the forefront of tying these technologies together and integrating them into one ecosystem of interoperable, collaborative frameworks where each piece contributes to a greater whole.  Enter the value of Smart Hospital strategy.

Smart Hospital strategy involves automating applications and processes via high power computing, machine learning and artificial intelligence to provide more predictive results for more cost efficient and expedient, secure patient care.

We are seeing these elements which contribute to Smart Hospital strategy across a broader landscape, shared commonly across provider, payer, pharma, and research facilities.  Smart Hospital Strategy aims to ingest and operationalize datasets to meet new criteria for value-based care, improving the patient and clinical experience, reducing per-capita cost of healthcare, and increase the health of populations.

Figure 1: IOT In Healthcare

IHI’s initial Triple Aim framework has spawned a more purposeful connected health objective translating to IT operational models that start with a digitized architecture, interconnecting assets from which MIoT was born.  MIoT, or the Medical Internet of Things, takes all medical devices into play and virtually, anything that can be assigned an IP address is fair game. (Figure 1)  This edge conversation draws enabling applications that take inventory of these assets and provides dashboards of transparency to what, when and how these assets are used, often monetizing them as an integral component of business operational health.

In recognition of the variety of these devices, whether within or outside the footprint of a brick-and-mortar hospital, a research institute or behemoth pharmaceutical company, the common consensus is that they all produce one common output: data. The problems stem from the multitude of manufacturers whose data output formats differ or whether their respectful metadata are unstructured or structured and lack standardization in terms of interoperability.

Digital Imaging and Communications in Medicine (DICOMR-TV) solves one interoperability challenge where different formats of real-time video from different equipment can be viewed simultaneously in a synchronized fashion – some of the technology behind Augmented Reality. Since these heterogenous devices neither transfer the corresponding medical metadata nor enable the synchronization of multiple videos, DICOM-RTV provides a solution to surgeons to receive metadata via streaming video during surgery regardless of sourced equipment.

To this point, these types of challenges are calling on new technology companies who are presenting applications that standardize these data formats to an exported dataset. This dataset can then be further analyzed to gain insightful objectives or to implement advanced data sharing techniques like blockchain, which will be covered later in this article.

Ultimately, analysis placed on a myriad of disparate datasets, regardless of format, can provide correlation or meaningfulness before a converged consolidated database is generated. Normally, this would take a team of people to access what data is meaningful but with the power of converged standardization and machine learning, huge datasets can be processed beyond the scope of human constraints. Reliable conversion and analysis of that data into deeper clinical insights saves time, resources, and money. What once took months, now takes merely hours.

Presently, healthcare organizations have a variety of applications tuned to optimize clinical processes and workflows over digitized infrastructure. The datacenter is transitioning to a highly centralized computing environment as a collaborative converged footprint, capable of collapsing a 30 rack, $11M datacenter investment into a single $500k stack. And we’re not just talking networking, servers and storage units, but platforms aimed at addressing the 4 V’s of Big Data: Velocity, Volume, Variety and Veracity.

Artificial intelligence is a huge part of that, especially since modern AI technologies are data driven.  The analysis of large volumes of complex data positioned on powerful high-power compute can now discover new relationships between the information entered and the desired results from the available data; and can adapt their reasoning based on new data. The process for arriving at insightful patient outcomes starts from simple data extraction from MIoT and is then taken through the journey of AI. (Figure 2)

Figure 2: Artificial Intelligence and Associated Methods

AI is a complex topic but when broken down to simple terms, the explanation offers a real understanding of its value.

First off, let’s understand what Machine Learning is. It all starts with an algorithm or a set of rules. Data is parsed or converted into a more readable, structured format that can be more easily understood.  Then, it starts the process of learning from that data. But before an algorithm can begin to learn from data, the data must be labeled as an objective knowledge base for what that piece of data is. The more details presented in terms of differentiating characteristics, the better an algorithm can be distinguished between classes.

Applied to healthcare, the more data you have about certain types of cancer the better a machine can identify what type of cancer any new incoming data sample most resembles. This can be applied to images, text, audio or other types of structured data to help create training sets that make training models or algorithms more impactful.

As a subset of ML, Deep Learning (DL) can learn what features in the dataset are relevant for a specific task. This is done by collecting a set of labeled datapoints and defining a neural network architecture suited for a given task.  This neural network consists of a set of weights and connections that are iteratively updated during a training process. The output of the training process is a trained network that can be applied on a new unseen data sample.  The artificial neural network (ANN) has learned what features are meaningful to differentiate, without having those rules manually defined.

The distinction between ML and DL is that in DL the descriptive features no longer need to be manually defined and can leverage both structured and unstructured data. Translated to healthcare, deep learning can predict a type of cancer in an image based on the training samples used to train the algorithm.  This training process generally needs a lot of data input, typically accelerated by GPUs.

While we have seen multiple iterations of digital transformation over the years in healthcare where one thing is certain: data. We have arrived at a turning point, due mostly to the introduction of meaningful use, in which patient data capture is almost completely digitized. This emergence can now generate the volumes of data required for building purposeful datasets for AI initiatives, delivering higher accuracy in predictive modeling. Healthcare organizations’ ability to harness this data for AI initiatives are quickly recognizing the economic payback. (Figure 3)

As AI is a continuous learning process and it is important to note that investing in scalable architecture that can grow with the expansion of AI initiatives resonates as a fundamental requirement. Healthcare organizations “will need to invest in technologies that not only integrate actionable AI insights into clinical workflow on the front end but also technologies that feed data into AI algorithms to generate insights.” (American Hospital Association)

The result? An invariable circulatory system of ingested data that can be pulsed to AI algorithms, gleaning continuous improvements in the clinical decision-making processes.

An area in which this process has shown applauding performances, is demonstrated through radiology applications. Solutions that provide scalable computing platforms that enables software developers to build and deploy medical imaging applications into hybrid (embedded, on-premise, or cloud) computing environments can create intelligent instruments and automate healthcare workflows. Augmenting radiology with artificial intelligence and deep learning is reinventing the way we visualize medical images.

For example, a Magnetic Resonance Imaging (MRI) scan, is an irreplaceable tool for diagnosing many medical conditions. It is time consuming, and in some cases, it requires a contrast agent to better visualize findings. Solutions like Nvidia Clara, accelerate MRI scans that can potentially be performed in a quarter of the time, and with less contrast, without sacrificing the quality of the image. “Most recently, ML techniques have focused on detecting COVID-19 infections in chest X-rays and CT scans [20][21]. Overall, sensitivity of CT scan has ranged from 57–100% for symptomatic and 46–100% for asymptomatic COVID‐19 patients [21].”

Another area that is gaining speed is in pathology. Typical digital image analysis tasks in diagnostic pathology involve segmentation, detection, and classification, as well as quantification and grading [44]. Due to the highly complex and multi-phased process in tissue sampling, one can expect some degradation of image quality through the process.

Frontiers in Medicine discusses how one of the largest single site pathology services in Europe NHS Greater Glasgow and Clyde, has begun proceedings to undergo full digitization.  “As the adoption of digital pathology becomes wider, automated image analysis of tissue morphology has the potential to further establish itself in pathology and ultimately decrease the workload of pathologists, reduce turnaround times for reporting, and standardize clinical practices. For example, known or novel biomarkers and histopathological features can be automatically quantified (412). Furthermore, deep learning techniques can be employed to recognize morphological patterns within the specimen for diagnostic and triaging purposes (1316).”

In applying Deep Learning to ML pipelines in pathology, half the battle is in defining pre-trained models. Like radiology, pathology offers up voluminous data. In ML though, while the volume of data poses a degree of effectiveness in outcomes, it can be said that at times, more weight worthy effectiveness is credited to the algorithm quality. Training datasets is probably one of the most time-consuming elements, so technology companies that offer up pre-trained models are worth their weight in gold.

The debate in terms of volume of data vs quality of algorithm can go on for days. Where there is little variance in metadata (high bias), regardless of volume of data, the algorithm applied to this dataset would obviously bear more fruit. Generally, the quality of the algorithm, obviously, improves as the dataset changes.

There are no shortages for the number of use cases supported by AI in Smart Hospital strategy. As we emerge from the effects of this pandemic, more considerations for broader AI adoption are being considered to avoid clinical staff burnout. Causes like alarm fatigue, ICU vital monitoring, or making life and death decisions. “And the economic impact is significant: physician burnout is conservatively estimated to cost the US healthcare system $4.6 billion per year in physician turnover and reduced clinical hours.” (Jeroen Tas, Philips, March 10, 2020)

The future and value of Smart Hospital systems lies within the ability to aggregate typically silo’d systems so AI algorithms can safely leverage medical history through the continuum of care against real-time vital reporting, especially if a patient has an underlying heart condition. In short, what this pandemic has taught us is that a pandemic can tax even the most efficient of healthcare system under a nation wide shortage of healthcare workers. Adopting point-of-care AI systems benefits not just the burden of limited resources, but also adds to the hospitals’ bottom line. Based on a report produced by Deloitte, hospitals were more likely to see higher revenue gains because of investing in AI and Smart Hospital strategy.

While we are on the precipice of embracing some of the aforementioned Smart Hospital strategies more widely across the healthcare continuum, the need for security and patient privacy, especially considering the uptick of Ransomware attacks, has never been greater. As more smart devices are introduced, higher scrutiny needs to be placed on protecting them and the data they generate.  Patient data has become even more valuable than credit card information. But it’s not always the data they’re after.

Medical instruments can be held hostage, rendering them unusable until the ransom is paid. Ransomware is an imminent threat and its consequences are far more than financial.  “A doctor at one affected target anonymously told Reuters that their hospital could not use some critical technologies, transfer sick patients or update electronic health records as officials dealt with the situation.  “We could still watch vitals and get imaging done, but all results were being communicated via paper only,” the doctor said.” (Kevin Joy, HealthTech, October, 2020)

Smart Hospital Strategy needs to incorporate stellar security practices to protect its investments and privacy of patients, but more importantly in the defense of saving lives.  All healthcare organizations have some security practices in place and medical devices are no exception. If attached to the network via a managing computer, cyber-attacks can still render the medical device hostage.

“Many device industry giants—including BD, Abbott, Siemens, Philips, Medtronic, Johnson & Johnson, Boston Scientific, and Strykerv—have pledged to publicly share vulnerability information in the event of a cybersecurity breach on their devices” (Nach Dave, IEEE Spectrum).  Nach further adds, “…, the Health Care & Public Sector Coordinating Councils issued a joint security plan that provides recommendations for managing the security of medical devices throughout the product lifecycle. Under this plan, health care providers and purchasers of connected medical devices would be able to remotely access a cybersecurity bill of materials (CBOM) that would list all commercial hardware and all software embedded in the device. The plan would also require device manufacturers to notify customers before ending technical support for older devices.”

There is credible information of an increased and imminent cybercrime threat” to U.S. hospitals and health care providers.“ (Cybersecurity and Infrastructure Security Agency, the Department of Health and Human Services and the FBI) Ransomware, Ryuk, has already impacted at least 5 U.S. hospitals in a single week and could potentially affect hundreds more. The hospital environment isn’t singled out. The pharmaceutical industry is an equal target.

“Cybersecurity incidents and ransomware attacks disrupt services and can also delay the progress towards drug development and delivery. Intellectual property, sensitive and personal information, are coveted by these hackers and during a pandemic, these threats are heightened.“

In March 2020, ExecuPharm, a company that provides clinical research support services for the pharmaceutical industry, was hit by Clop ransomware. The ransomware group behind the attack published the sensitive data stolen from the company’s server.” (WhiteHat Security, Shweta Khare, November, 2020)

Cybersecurity solutions in software as well as hardware are constantly evolving with each threat. With the advent of SmartNICs, anti-malware solutions are integrated as smart-agents on data processing units so core processes remain untaxed. Security applications that include intrusion detection and prevention, while operating on the bare-metal server, permits the hypervisor to move from the x86 CPU to run instead on the DPU, embedded arm CPU cores, freeing up the host x86 CPU to run applications, running in a fully isolated DPU domain. The result is better application efficiency and isolation between the host CPU application and DPU management domains.

When it comes to clinical research, isolation is key – whether it’s securing cloud container services or implementing perimeter detection solutions that constantly monitor code changes to web applications, the collaborative nature of research demands reliable, secure and scalable access, as healthcare organizations have grown more receptive to cloud data repository and web-based applications, especially where EHR is concerned. Now that patient data as well as intellectual property from pharmaceutical industry is digitized, there is a lot more to lose but exponentially, more to gain. Enter AI over Blockchain.

Blockchain technology is making moves to provide controlled and secure access to data.  Just 5 years ago, this technology was regarded as merely hype but the advancement of this technology along with cyber security risks has made giant leaps in the foray of secured ML.

As opposed to a public blockchain commonly represented as Bitcoin, a private blockchain is an invitation-only network controlled by a single authority or by a set of rules (smart contract), hence given the name permission-based blockchain.  Participating entities invited to the private chain can have variable access. Some can be granted permissions to access certain types of data with limited functions. Everything else is off limits.

Smart contracts can operate documents and data stored outside of the blockchain, which is not possible for public blockchain platforms.  Through permissible blockchain every piece of data is heavily encrypted, so one can appreciate the importance of accurate and secure datasets to build more effective ML models. There have been a multitude of books, articles, consortiums, and white papers written about this practical use.

When it comes to security, healthcare organizations are exploring its valuable application and pairing its capabilities with AI.  It introduces a methodology of adopting a whole new layer of security and accountability over which specially applied algorithms for AI driven applications can be utilized.  Use cases like supply chain management, anti-counterfeit drug production and tracking, clinical trials or federated research are benefitting from blockchain. Medical devices operating under blockchain can now be tracked from manufacturing, to system repairs, to operating code modifications. For the life and integrity of the machine, having a transparent view of its lifecycle is critical when interfacing with patient treatment, including implantable pacemakers, to reduce security vulnerabilities.  Blockchain’s inherent capability of maintaining data ownership enforces accountability for patient care.

“Strong cryptographic security guarantees and the distributed nature of governance means the risk of unauthorized write operations or successful attacks against the blockchain are less likely versus a single point of failure or single source of truth such as a central database, and thus protects the data stored and transmitted by the blockchain from unauthorized manipulation or alternation “.  (Blockchain in Healthcare, Metcalf, Bass, Hooper, Cahana and Dhillon)

This technology can be just as valuable on smaller scales within a single domain. During the pandemic, Personal Protective Equipment (PPE) tracking due to supply shortage became a huge concern. By enabling a distributed ledger that introduced decentralized control, security, and traceability, organizations were benefited to having more predictability in supply, where items like mask and glove quantities met demand needs.

MeriTalk’s Kati Malone recently wrote an article discussing the benefit of a program called Accelerate, producing $30M in savings appreciated by the HHS as a result of leveraging blockchain. “HHS Deputy Secretary Eric Hargan shared an example of a behind-the-scenes technological advancement unrelated to Accelerate that he said has also generated billions of dollars of “regulatory savings” through the use of AI-assisted analyses and other initiatives.”

Intermountain Health has also benefitted from blockchain and AI with a $90M savings. (Andrea Tinianow, September 23, 2019) The Forbes article adds, “Intermountain is using blockchain-based technology coupled with artificial intelligence to identify waste in its massive healthcare system, creating better outcomes for patients, and significant savings all around.” Additionally, organizations like Mayo Clinic, St. Joseph’s Health, and Novartis are the recipients of its merits.

The immutable nature of blockchain ensures quality control where distribution safety adheres to strict CDC guidelines. From the source of manufacturing, these auditable time-stamped transactions are shared across the chain keeping clinical staff safe from faulty products, like PPE. Blockchain is quickly becoming an industry technological revolution that is providing solutions to a multitude of use cases across all industries.

“Global blockchain in healthcare market is forecast to grow at a CAGR more than 70% during 2020-2027, on account of increasing instances of data breaches and surging adoption of blockchain technology in healthcare and pharma industries.” (Business Wire, March 13, 2020) The blockchain market in healthcare was valued at USD 2.12 billion in 2020, and is expected to reach USD 3.49 billion by 2026, with a CAGR of 8.7% during the forecast period, 2021 – 2026. (Mordor Intelligence)

Summary

We have discovered more ways to network various types of instruments for a multitude of use cases supporting the Smart Hospital Strategy. Further, we look to extracting critical data produced by these instruments to gain insight into clinical and business operations.

Introduce Deep Learning where a type of artificial intelligence leveraging computing programming learns information without human intervention. AI leverages machine learning algorithms and additional software to mimic human cognitive functions in the analysis, comprehension, and presentation of medical data. “Machine learning algorithms and their ability to synthesize highly complex datasets may be able to illuminate new options for targeting therapies to an individual’s unique genetic makeup” (Jennifer Bresnick, Healthcare Analytics).

In effect, AI can expedite patient care accurately and swiftly while saving healthcare costs. It is also important for healthcare organizations to work with their device manufacturer on security plans, often working with R&D to incorporate mechanisms such as advanced SmartNICS whose onboard data processing units take the responsibility of running anti-ransomware agents to provide a first line of defense. And to further aid in secure data sharing across large organizations, blockchain is quickly becoming an emerging technology seen more pervasively across the healthcare spectrum in appreciation for its ability to track, audit and protect valuable data while being able to contribute higher instances of effective machine learning models.

Smart Hospital strategy can help leverage data as the most prized asset, progressing checkerboarded, silo’d use cases to moves where each technology begins to reveal a holistic plan in the chess game of industry differentiation, market share and leader of quality health.

About the author

Guest Author

1 Comment

  • An extremely insightful, informative and realistic view of such an important topic! Ms Millin has produced a piece of work that deserves congratulations for its clarity, giving us an easily understood picture of how healthcare has progressed through technological improvements and how health organizations can benefit and improve our lives further through the use of AI, ML and DL, Blockchain etc. an excellent and very useful article.

Click here to post a comment
   

Categories