Watson Joins the Digital Health Record Mix
By Sandra Nunn, MA, RHIA, CHP
Vol. 17 No. 12 P. 22
A new medical imaging collaborative seeks to leverage IBM's Watson for better patient data.
The idea of a longitudinal patient record (LPR) is not new. Once used to describe a patient's medical records existing in multiple locations along the continuum of care, including ambulatory, acute, subacute, rehabilitation, home care, nursing home, and hospice, electronic health information vendors have breathed new life and meaning into the term. The new LPR describes multisource retrieval of patient data, including the images and documents generated by radiologists and other care providers, to isolate previously undetectable disease anomalies through the sifting of thousands and even millions of images and documents consolidated in image libraries housed in the cloud.
A recent presentation by Allscripts' Harwant Sethi, director of sales engineering for population health, featured the "Technologically-Enabled Longitudinal Patient Record: The Vision: A Closed Loop Healthcare Platform." The renovated LPR concept proves that the goal to relate patient information over its lifecycle from multiple sources is now potentially obtainable.
Recognizing the enormous potential of the LPR concept, Tanveer Syeda-Mahmood, PhD, senior manager and chief scientist for the Medical Sieve Radiology Grand Challenge Project at IBM, won the right to lead IBM's Grand Challenge project through her idea to develop the medical sieve technology. According to information provided by Sabanci University in Turkey, she proposed that IBM devote research to the following key problems:
• "disease understanding by diagnostically interpreting medical imaging datasets;
• modeling and assembly of top-down and bottom-up clinical knowledge and its use in clinical reasoning; and
• intelligent filtering or summarization of content for presentation to clinicians."
Beyond the technological research required to fulfill the above requirements, an additional vision was required: the formation of a consortium of entities willing to interact with one another in terms of sharing radiology images and reports based on an agreed-upon clinical vocabulary and development of a community of knowledge regarding electronic storage management in the cloud. The electronic storage would need to be able to house and manage large image files (X-ray, ultrasound, CT, MRI, PET, and clinical text) from multiple entities over the span of years.
Still in development, Medical Sieve "is an image-guided informatics system that acts as a medical sieve filtering the essential clinical information physicians need to know about the patient for diagnosis and treatment planning. The system gathers clinical data about the patient from a variety of enterprise systems in hospitals including the EMR, pharmacy, labs, ADT [admission, discharge, and transfer], and radiology/cardiology PACS systems using HL7 [Health Level Seven International] and DICOM adapters. It then uses sophisticated medical text and image processing, pattern recognition, and machine learning techniques guided by advanced clinical knowledge to process clinical data about the patient to extract meaningful summaries indicating the anomalies," according to information provided by IBM.
Physicians on the receiving end get advanced studies that point out the important anomalies that Watson's huge processing powers can detect. Advantages for radiologists and other physicians include the assurance that they are aware of all data relevant to a particular patient, they will be able to review statistical comparisons to similar patient groups, coincidental diagnoses will be surfaced, patients can be triaged, and the potential for automated report generation exists. In addition, the medical sieve technology lends itself to predictive analytics and demonstration of protocol adherence.
Thinking Like a Radiologist
José Morey, MD, a senior medical scientist for IBM Research, a visiting assistant professor in the department of radiology and medical imaging at the University of Virginia, and a radiologist at the VA in Norfolk, Virginia, enthusiastically relates his experiences over the last few years in training Watson to "see." Describing the IBM challenge projects as "moonshots," Morey, one of eight practicing physicians who are helping out with the Medical Sieve project, talks about "smart machines who don't really think like humans but need to know how humans think and what they know." Morey's job, in conjunction with a wide range of experts in other domains, is to "teach Watson to think like a radiologist." He describes teaching the computer the concept of "hand" by showing it multiple iterations of hand images and concepts of "hand."
Morey has developed a methodology that he executes in thirds: training of the machine, testing of the training, and review of the quality of the output. Watson's training modules are integrated with reasoning components that include patient history, medication, and, if available, DNA analysis. The integration of all of this data yields a differential diagnosis. According to Morey, the goal is not to replace the radiologist or physician but to augment their abilities and productivity, which would give them more time to see more patients.
Morey is also an idealist. He believes in the "democratization of medicine no matter where you are or who you are;" through the power of trained entities like Watson, as well as web and cloud resources, some level of care could be made available no matter where a patient might be in the world. With Watson, Morey says that he "can help more than 100 to 200 people a day, people I will never see nor will they see me."
Morey looks at the LPR from a computer science perspective, ie, there is an ever-increasing volume of data, and the algorithms developed to handle those data are getting better and better. He says the Affordable Care Act has the potential to create health record retention that would be ad infinitum. The Medical Sieve project steps into cognitive computing based on the enormous amount and types of data available and the algorithms that will be able to say "how, why, and when things are going to happen." Morey is employing deep learning, which is a branch of machine learning based on a set of algorithms, to train Watson. Deep learning attempts to model high-level abstractions in data by using graphs that allow for multiple processing layers and the representation of data.
Discussing Watson Oncology, Morey explains that the Medical Sieve is a tool that is already partially commercialized. Through the VA, Vice President Biden has decided to partner his resources from the federal cancer cure efforts to support cognitive health, and Memorial Sloan Kettering Cancer Center in New York City is involved in genomic analysis and more specific, deeper learning from the resulting data. Relating this initiative to the LPR concept, Morey sees it as a chance to "leverage more data for better reasoning tools. Radiologists will need to be health care savvy but also much more IT savvy; it's better to be at the table than on the menu."
Above and beyond a fluency in EHR interactions, future radiologists will need to understand two branches of computer science: computer vision and machine learning. It is radiologists who will define and refine the content of images and choose the attributes most effective to classify it. Morey emphasizes that the algorithms he creates are cognitive, ie, designed for the computer to teach itself. Watson is constantly learning through correcting itself to get closer to better results.
Seeking a competitive edge, IBM is working to make cognitive computing through the Watson Health Cloud, a cloud-based platform, commercially available to diverse customers. Watson's ability to analyze enormous volumes of data, handle complex questions posed in natural language, and propose evidence-based solutions is drawing on the contributions of clinical information from multiple radiologists, other clinicians, researchers, and patients' wearable devices. This constantly deepening lake of data will be acted upon by Watson, which continuously learns from the increasing store of data, from its own experiences, and from other knowledge over time. Due to its deep bench of applications, IBM can draw on master data management experience in managing patient indices to support integration of multiple patient files, with high integrity, into the Watson Health Cloud.
In early June 2016, IBM Watson Health launched a medical imaging collaborative to support data integration and sharing within the Watson Health Cloud. Morey notes the important role of the collaborative and the need for standards building on the HL7 and DICOM existing measures; individual members of the collaborative often have made internal data integrity efforts to ensure data integrity within their organizations, which will ease their efforts to feed deidentified patient information into the larger network.
The collaborative includes 16 other health systems including academic medical centers, ambulatory radiology providers, and imaging tech companies. Sheridan Healthcare, the largest hospital-based radiology services provider in the country and a specialist in the provision of teleradiology services, became one of the first members. Sheridan and other collaborative members hope to gain insights from "previously 'invisible' unstructured imaging data and combine them with a broad variety of data" to customize physicians' care decisions and make them more relevant to the specific patient, according to press materials.
A fortuitous byproduct of this broad-spectrum data gathering will be the construction, in coordination with Watson Health and other collaborative members, of a body of knowledge that can benefit broader patient populations and support efforts to develop interoperability with other health care entities. Additional dimension and range were added to the Watson Health collaborative when Agfa HealthCare became a participant in late June 2016. Agfa has "over a century of healthcare experience and has been a pioneer on the healthcare IT market" and "designs, develops, and delivers state-of-the-art systems for capturing, managing, and processing diagnostic images and clinical/administrative information for healthcare facilities," according to the company's website.
Members of the collaborative will provide clinician-generated clinical knowledge to the Watson Health repository. This knowledge comprises patient histories and physicals, diagnostic examinations and lab reports, anatomical and physiological reports, pathology reports, vital sign data, measurements and clinical guidelines, procedure and treatment documentation, past medical histories, and even signs and symptoms information. In addition, the repository will have the latest authoritative knowledge sources from classification and taxonomy systems like the Systematized Nomenclature of Medicine and Medical Subject Headings. Through natural language processing, Watson will also contribute pertinent journal articles and textbook sources. Watson's text mining capabilities can extract other potential concepts and relationships as well and direct them to physicians to contribute to clinical decision support (CDS).
If You Build It …
On a more practical financial and operational level, members of the collaborative are hoping to experience increases in productivity and reductions in cost due to the additional harnessing of technological power. Anne Le Grand, vice president of imaging for Watson Health, says in a press release that "through IBM's medical imaging collaborative, Watson may create opportunities for radiologists to extract greater insights and value from imaging data while better managing costs." From a practice management perspective, collaborative members and their radiologists will be able to monitor performance against industry benchmarks, address workflow issues—including evening out the distribution of work among providers and departments—minimize inefficiencies, track pay for performance indicators, and justify additional budgetary resources based on automatically generated data.
Additionally, Medical Sieve perceptions are being applied to CDS, which has historically been linked to feeds of structured data. Watson offers the growing facility to access previously untapped information locked away in unstructured images and narrative reports. Training Watson to "see" anomalies in moving image scans by teaching it what a normal visualization is gives the computer semantic understanding of what is considered a normal study. Potential CDS benefits include the ability to better manage the radiation doses that patients receive, the ability to act on suggested enhancements to drug efficacy, quicker response to the need to adjust utilization of imaging resources, and possibly the ability to conduct clinical trials with vaster populations in diminished time frames.
Understanding that Watson Health would need a very big space to house all the information assets likely to be generated by the multiple participants in its consortium, IBM acquired Merge Healthcare late in 2015. Representing IBM Media Relations, Christine Douglass says in a press release, "Merge's technology platforms are used at more than 7,500 US health care sites, as well as many of the world's leading clinical research institutes and pharmaceutical firms to manage a growing body of medical images." Merge's acquisition serves to bolster the already existing "315 billion data points in the Watson Health Cloud, including lab results, electronic health records, genomic tests, clinical studies, and other health-related data sources," according to press materials.
The gigantic structured and unstructured data repository that IBM is building to provide Watson its viewing space can certainly be intimidating for the everyday provider. In "A Generalized Framework for Medical Image Classification and Recognition," published in 2015 in the IBM Journal of Research and Development, Albedini and colleagues declare that "medical image data has been growing by 20% to 40% every year, whereas the number of physicians per capita in the United States has remained relatively flat since the 1990s." With the constantly expanding volume of images flowing into the Watson Cloud, a means of automatic classification and categorization of medical images has become imperative for physicians to handle growing workloads. The paper proposes a visual machine-learning framework for classification of medical images. The authors have come together to begin to build a standardized, more universal way to organize images across several domains including type of modality, body region, view, and disease. The Medical Sieve project is one of many under way by IBM and other health-engaged organizations that build on the LPR concept and lend hope to early disease detection and, in time, disease elimination.— Sandra Nunn, MA, RHIA, CHP, is a freelance writer and principal of KAMC Consulting in Albuquerque, New Mexico.