Digital Healthcare: Challenges, Opportunities to Innovate, and… Leave a comment
Image courtesy of Patrick Bangert
A significant portion of society is familiar with digital transformation, from experiencing telehealth during the pandemic to the many wearables available on the market today. And while a seemingly more efficient approach, this foray into digital healthcare is hampered by disparate and often unreliable data, interoperability challenges, among other obstacles.
Two renowned industry experts shed light on the many opportunities and challenges associated with the maturation of digital healthcare at IME West, an advanced manufacturing trade show, held this past April in a highly sought-after session exploring AI, big data analytics, and edge computing:
- Nandha Kumar Balasubramaniam, Director of Healthcare Genomics & Medical Device Strategy Healthcare Global Business Unit at Oracle Health
- Patrick Bangert, VP of AI at Samsung SDSA
Given the complexity and ever-progressing nature of this topic, I sat down with Nandha and Patrick to delve deeper.
Adrienne: For those unable to attend, could you tell us about the session you led alongside Patrick Bangert?
Nandha:Patrick and I were given a Center Stage position in a fireside chat format to discuss our views on the latest trends and challenges in Digital Health. It was exciting to share our expert opinion on how Artificial Intelligence (AI), Big Data, Edge Computing, and Healthcare Genomics transform NextGen Healthcare. We had an energized audience who were motivated for face-to-face interaction to learn about these key trends, voice their opinions, and enable us to validate our findings.
We highlighted the key trends influencing healthcare, beginning with the move toward Personalized Medicine/Precision Health which is increasingly relying on genomics data to derive meaningful insights. This is based on the rapid advancements in gene sequencing technology, expansion of Omics technologies, increased affordability, and combining genomics with EHR data. The focus on population sequencing programs, innovations necessitated by the recent pandemic, increased reliance on big data analytics while maintaining data security and privacy are key to better patient outcomes.
Moreover, we also discussed the specific applications of Artificial Intelligence (AI) and the challenges pertaining to Digital Health. AI is being increasingly applied for image analysis. I highlighted the use of Deep Learning algorithms by the Ellison Institute for Transformative Medicine to identify breast cancer tumor markers from histology stained slides. The outcome from this application was shown to be equal to or better than the traditional methodology of examining the stained slides by microscopic image analysis. Simultaneously, I also highlighted the key challenges that need to be given due consideration: the need for high-powered computing that is expensive, poor quality and quantity of data sets combined with data bias, data silos, and effective governance. Then comes the need to understand the process employed in developing the algorithms combined with an all-inclusive diverse team.
Patrick:Digital health will revolutionize the healthcare industry and patients’ healthcare experience. In my opinion, AI will be the mechanism that makes this happen. The revolution itself consists of three major elements: (1) A significantly reduced time delay between a health exam and its actionable results, (2) much more doctor time spent facing the patient than dealing with data entry and data analyst activities, and (3) better medical outcomes due to faster treatment and more accurate diagnoses. A side effect of this will be lower costs.
There are two entry barriers, and they are creating sizeable high-quality datasets for AI to learn from and generating trust among doctors and patients. The cost and effort of creating a good quality dataset are largely comprised of the cost of labeling data. It may take one hour for a specialist to annotate a single X-Ray, MRI, or CT scan and perhaps four hours to annotate a pathology whole slide image. To avoid human error, it’s typical for three experts to label each image. As relevant datasets need hundreds of thousands of images, the investment is immense. Automating and quality checking this process is a significant enabler of healthcare AI.
Trust in AI is not where it should be. People often point to the Terminator or basic failures seen in the media. It is hard to tell a well-made model from a poor one – certainly harder than telling a toy car from a luxury car. But that is the right analogy. With the proper safeguards, responsibly made AI models that explain their outputs are possible and can serve as decision-making tools for doctors.
Adrienne: Let’s dive into this a little deeper. From your perspective, what is the number one issue in Digital Healthcare that needs to be tackled immediately?
Nandha: One of the key research findings in digital health has mentioned that 10-15 connected devices are used per hospital bed and that the average clinical trial generates three million data points. This, in turn, highlights the need for digital interoperability, which I feel is one of the key challenges that need to be addressed in digital healthcare. Addressing this essential need will, increase the overall patient experience, clinical efficiency, lower healthcare costs, and offer better outcomes for the patient.
Interoperability challenges arise from poor data quality stemming from data errors and misrepresentation, lack of data standardization, and the incomplete adoption of existing standards. Personal Connected Health Alliance and IHE International are stressing the importance of implementing interoperability standards and other areas of digital health. When multiple devices are employed to monitor patient vitals, the risks arise from device errors, network failures, security threats, and the ease of inter-device communication. When these medical devices originate from multiple vendors, due consideration should be given to the agreements signed with the vendors on liability and indemnification when there are system failures and security breaches.
Patrick:I believe it is the Electronic Health Record (EHR), also known as the electronic medical record. These systems were created with the promise of saving doctors’ time and keeping all relevant information in one place. They do none of these things. Doctors are using more time than before feeding the beast, often having to input a piece of information multiple times. Much of the information is irrelevant, and not all relevant information is necessarily there. Finally, EHRs are not one place at all. Every hospital system has its own and making them talk to each other can seem like a UN conference.
We need EHR v2.0 with standardized input and output to make all systems interoperable. They need intuitive, easy-to-use interfaces for doctors to ergonomically and efficiently enter information – ideally, this information should be automatically extracted from the verbal conversation between doctor and patient by AI. Finally, the information presented to a doctor reviewing a case should be selected based on what is relevant to the current challenge by AI.
Adrienne: Where are the unique opportunities for innovation in Digital Healthcare? (Data, Privacy & Security, Compute, Governance, to name a few.)
Nandha: Data, Data, Data – are the three magic words in digital healthcare, for when the data is of high quality, it enables deriving meaningful insights for better patient outcomes. On the other hand, all hell breaks loose when data is of poor quality. Processing a large volume of data necessitates the innovation in Cloud infrastructure, High-Performance Compute, and autonomous databasing, to name a few.
Innovations in Cloud computing, such as Gen 2 Cloud, for example, from Oracle, enable a medical device company to run any workload to outgun a data center’s performance, control, and governance. This can be done at the needed scale, flexibility, and lower cost. High-Performance Computing (HPC) is another innovation that makes a difference in handling complex and voluminous data, along with scalable storage and low latency. Finally, the autonomous database from Oracle introduced in 2018 redefined the concept of database management. Leveraging machine learning reduces the error rates, the need for human labor, and offers a high degree of reliability, security, and operational efficiency. Also, it provides the ability to apply patches and upgrades without incurring downtime and the need to fine-tune the engine to achieve optimal performance to match the workloads.
Patrick:Many of the ingredients exist today in other fields such as privacy, security, data governance, and so on. A challenge particularly acute to healthcare and not yet adequately solved by anyone is explainability. This is and will increasingly be a major frontier research area. AI models today simply provide an output. Put in an MRI; the AI will say “cancer.” That may be correct, but it’s not helpful. What everyone wants is an explanation that a human being – perhaps only an expert – can comprehend and agree with. Ideally, this explanation uses higher-level concepts and ultimately instills trust in the final answer that the AI provided. I believe this is the crucial obstacle remaining.
Adrienne: Do you have any regional success stories or learnings in Digital healthcare that can be propagated to other parts of the globe?
Nandha: Yes, I can highlight Oracle’s partnership with Xpert.AI under the new start-up initiative. Through this partnership, Xpert.AI was able to capitalize on Oracle Cloud Infrastructure to bring together vast amounts of unstructured data from brain scans (images for pattern recognition in real-time. This involved capitalizing on technologies such as Natural language Processing (NLP), the Internet of Things ((IoT), Flash GPUs, and the Oracle Cloud Infrastructure (OCI) to enhance cognition in patients suffering from Epileptic seizures. Compared to the traditional model, which is expensive, and needing the patient to be at the hospital, the Xpert.AI solution can perform diagnosis of seizures using Remote Patient Monitoring (RPM) with the patient is at home. Deep Learning (DL) algorithms can track the origin of the seizures and track it across the brain, and the data is sent to the clinician for diagnosis of the seizures. Hence, the combination of Deep Learning algorithms and Remote Patient Monitoring can be used to predict seizures and is a game-changer for 65 million patients globally.
Patrick:Samsung SDS has a partnership with the City of Hope, a major cancer center in Los Angeles. We work together on digital pathology, where whole slide images are created from biopsy slides. These high-resolution images contain thousands of details that must be found, analyzed, and counted to produce the final result. We are providing the AI processing; together we are making models that will eventually enable hospitals to reduce the six-week wait from biopsy to diagnosis to almost nothing at all, with a higher accuracy rating than before.
