By Dan Harvey
Vol. 20 No. 12 P. 14
Integrated diagnostics promises to improve diagnosis and treatment.
One of the notable sessions at RSNA 2018 explored the concept of integrated diagnostics, which involves developing a closer relationship among elements that include radiology, pathology, and genomics. According to the presenters of the session, integrated diagnostics can substantially impact radiology workflow, improve diagnoses and patient care, reduce costs for both health care enterprises and patients, and increase patients’ satisfaction with their experience. The topic will again be spotlighted at RSNA 2019.
Integrated diagnostics is a wide-ranging region of research that considers how tests are ordered and performed as well as how to effectively combine tools such as AI and data analytics. The goal is to achieve tighter integration and communication among the three aforementioned disciplines. Integrated diagnostics has been described as a one-stop setting for testing and additional procedures and a “computational revolution” that should help break down the silos that separate specialties such as radiology, pathology, and genomics. Although interdependence has existed between these disciplines, it has become recognized that a more defined and closer collaboration is necessary. It’s important to note that the move toward this concept involves a move away from the analog and into the digital.
A Researcher’s Perspective
Anant Madabhushi, PhD, FAIMBE, FIEEE, a research health scientist who serves as the director of the Center for Computational Imaging and Personalized Diagnostics in the departments of biomedical engineering, urology, radiology, pathology, radiation oncology, electrical engineering and computational science, and general medical sciences at Case Western Reserve University in Cleveland, who will give a presentation on the topic at RSNA 2019, provides a researcher’s perspective. He sees integrated diagnostics as an opportunity to employ radiology and pathology in a unified way to improve diagnosis and prognosis, treatment planning, and treatment management.
“I am not a clinician, so my focus is research centric,” Madabhushi says. “I work in the areas of artificial intelligence and computation, and I apply those to both radiology and pathology.”
He describes how the process currently works in many cases: There may be a patient who underwent imaging procedures, such as MRI or CT, and biopsies. The pathologist looks at the biopsies, and the radiologist looks at the images. From there, a diagnosis is rendered.
“The goal is to render the best possible diagnosis, prognosis, and subsequent treatment strategy. So, it behooves us to look at how to cohesively integrate all gathered information,” Madabhushi says. “On the research side, we’re looking at patterns in imaging scans to best determine risk stratification. Similarly, within pathology, analysis of AI and radiomics determines the degree of disease aggressiveness and predicts treatment results.”
For example, early-stage lung cancer presents a question of which patients need chemotherapy after surgery. Typically, patients have had both imaging scans and surgery. To best determine whether a patient needs chemotherapy, the clinician needs to interpret the scans. At the same time, the tissue images need to be analyzed.
“What we’re doing now [with integrated diagnostics] is taking patterns mined from CT scans and combining the data with patterns mined from tissue images,” Madabhushi says. This, he says, creates a unified predictor of outcome, helping clinicians better determine which patients will benefit from adjuvant chemotherapy following surgery.
A similar situation can occur with prostate cancer patients.
“Very often, the question is whether the cancer requires surgery or radical therapy or active surveillance,” Madabhushi says. “Patients get MRI scans, so here is an opportunity to determine patterns on the scans and look at patterns of pathology, combine them, and then determine the best treatment strategy. Some patients may need surgery, but other patients with less aggressive disease may just require active surveillance.”
Combining information from pathology and radiology provides a better option on how to manage a patient’s disease, he says. “It goes beyond the silo approach where you look at radiology or you look at pathology,” Madabhushi says. “The silo approach misses a great deal.”
A New Direction
Madabhushi became intrigued by the integrative diagnostics concept during the graduate portion of his education at the University of Pennsylvania in Philadelphia. His PhD work involved biological engineering—specifically, the deployment of AI with radiologic imaging, which occurred when he connected with university pathologists who furthered his knowledge about AI. The first condition he looked at was prostate cancer.
“The disease is especially interesting because it provides the opportunity to do a presurgical scan,” he says.
In his subsequent graduate research, he collaborated with like-minded individuals to combine pathology—tissue from a biopsy—with MRI and began to map pathology patterns with those in the MRI scans.
“It allowed us to ascertain on the MRI the location of disease, and pathology provided the ability to look at the disease on a microscopic scale,” Madabhushi says. “With that combination, the pathologist was able to best identify the exact location of the disease. With computational tools, we could combine the biological specimen with the preoperative MRI data, and with that we could begin to develop AI algorithms on the scans.”
The addition of genomics enabled multiscale information, which, Madabhushi says, allowed for better categorization of the disease. “Genomics enabled me to identify the genes and the proteins that were associated with the disease,” he says. Collation and association became much clearer.
“I started with prostate and took it to lung cancer, brain cancer, and colorectal cancer,” Madabhushi says. “Connecting all of these informational branches provided a comprehensive portrait of a disease.”
At RSNA 2019, Madabhushi will present “Radio-Patho-Genomics: Computationally Integrating Disease Specific Features Across Scales,” wherein he will describe what he has learned and continues to learn about integrated diagnostics, focus on opportunities represented by categorization of measurement scales, and discuss why patterns found within MRI scans correspond with pathology information.
“This information can reveal who will have the best response to chemotherapy,” Madabhushi says. “By correlating information from radiology and pathology, we’re able to provide an intuitive basis as to why AI algorithms work the way the work—why they’re prognostic, why they’re predictive, and why they’re of therapeutic benefit.”
A Clinician’s Perspective
Mitchell Schnall, MD, the Eugene P. Pendergrass Professor of Radiology and chairman of the department of radiology at the Perelman School of Medicine at the University of Pennsylvania, provides a clinician’s perspective. Integrative diagnostics can mean different things for different people, he says. He describes the concept as a program or process that creates a previously nonexistent infrastructure. This process would involve knowledge and workflow that bring together diagnostic modalities under a single roof, which he believes would open the door for clinicians to start thinking about diagnostics in a more holistic fashion.
Clinicians have several diagnostic streams that provide information. These include laboratory tests, histopathology, and radiology as well as specialty-specific tests such as ECGs and electroencephalograms. All elements involve different workflows for ordering, results, and storage, but there’s no process for systematic integration of information. Everything is in silos.
“The way we think about integrated diagnostics involves several components,” Schnall explains. “There is a pure clinician workflow component. We think about the creation of a single place. This gives us the ability to determine what is most necessary. We used to think about testing in a radiology sense, a pathology sense, or a laboratory sense. But these are interrelated. Under certain laboratory conditions, certain image tests may seem necessary; some may appear to be unnecessary. So that leads us into workflow and knowledge base. How we should consider work ordering in the most integrated fashion?”
This leads to consideration of results from measurable places. A clinician needs to know how to put them all together. Having a single process that provides for expediting results and accessing them in a single place will improve the experience for not only clinicians but also patients.
“Patients scheduled for tests have to check in, get blood drawn, then go for radiology tests, all at different places,” Schnall says. “Can’t we put it all together so that the patient need only have one experience?”
From an operational workflow perspective, he says, integration creates substantial value. “We think about bringing together modern-world data that involve data analytics in ways that allow us to consider the patient unit and how we can achieve diagnosis as quickly as possible, considering all the kinds of tests that need to be ordered,” he says.
A pilot study that Schnall and colleagues conducted involved CT check-in at an outpatient area. Many patients needed blood tests, and they received them while awaiting their turn at a CT scanning machine.
“It saved the patient from having to go to another place and having to wait,” Schnall reports. “Everything can be done at the same time and place.”
Echoing Madabhushi’s concern, Schnall says that the two largest disciplines currently testing are pathology and radiology. “That compelled us to approach the two disciplines together and build a structure that allows for integration of different parts of testing done within different parts of an enterprise,” he explains. “This doesn’t mean a fundamental change of disciplines or a take-away of business from anybody. Rather, this is about collaboration and how we combine different diagnostics streams to improve patient diagnosis and, ultimately, patient care.”
Collaboration Is Key
The need for collaboration will only become more intense as molecular diagnostics performed on blood samples come into wider use. “We need to know how we use that in combination with imaging studies,” Schnall says. “What comes first? What comes second? We are going to need to be collaborating very closely to figure out the correct answers.”
Collaboration is achievable, he says, because radiology and pathology possess complementary expertise. Radiology specializes in biology—anatomy and pathology. Pathologists are exceptional at categorizing and creating structure around information. “Radiology is very good when it comes to IT, pushing pixels around, which proves very informative for medicine,” Schnall says.
He adds that diagnostics should be considered holistically—eg, via AI, computation, or data analysis. “Those are some of the tools that can make this happen,” he says.
Like Madabhushi, Schnall will deliver a presentation focused on integrative diagnostics at RSNA 2019. “The Path to Integrated Diagnostics” will delineate how Schnall and colleagues approach the process and the complexities involved. The presentation will also address how to deal with potential roadblocks.
“As you work together with disciplines, the faculty are going to feel very threatened,” Schnall reveals. “So, we started in a tactical way, working on both the front end and back end.”
There are other specific questions his presentation will focus on. These include how to accomplish a holistic approach to the ordering of tests and the previously unrecognized opportunities that may present themselves through collaboration and parallel workflow. Answers could enable development of a data model through analytics and lead toward a viable integration of surgical pathology and radiology practice. “[But] we’re not near that yet,” Schnall concedes.
One potential example that Schnall cites involves placing a breast imaging radiologist with a breast pathologist in the same place. He says there’s no need to change anything in the disciplines; let them work as they have worked.
“That way, they can ask each other questions,” he says. “Pathologists look at images all of the time, but they don’t really know enough radiology for that discipline to be as informative to them as it could be. Conversely, radiologists look at pathology results but don’t know enough about that discipline. If we can put it all in the same space, we can ask questions of each other. My guess is that this will reduce much of the uncertainty that comes with diagnostics.”
Although integrative diagnostics sounds promising, challenges remain, and some of them have nothing to do with technology or existing infrastructure; they involve the wiring of the brain.
“Frankly, the biggest challenge is culture,” Schnall says. “How do we decide, and how do we have the political will to say, ‘Get out of your comfort zone?’ How do we get out of our individual silos? How can we work together in comfort?”
He believes strong-minded people willing to form solid partnerships among departments will be needed. “We need strong leadership at the highest enterprise levels, among those who can recognize a clear demonstration of the value,” he says.
Madabhushi’s thinking runs along the same lines. He says the main challenge is “changing mind set.” Training also comes into play. “You train in radiology or pathology, not both,” he notes. But he sees a path forward.
“AI can disrupt medicine, but in a good way,” Madabhushi says. “It’s going to have to involve a lot of enthusiasm—that’s the best way I can put it—from the different specialties. We need to rethink the way we do things, which involves training and making specialties give up their silos with a kinetic energy. This is not going to be at all easy. We’re talking about changing arrangements that have existed for a long time.”
He believes that this can be accomplished through appreciation and discussion, which should provide a positive force toward change.
“If we keep plugging away, we will get beyond the silos and into truly integrated diagnostics because, at the end of the day, we strive for the best outcome,” Madabhushi says. “Those of us who work in biomedical science, clinical science, and clinical research can figure out a way to leverage all the information in the most comprehensive way possible to render the best prognosis and treatment strategy.”
— Dan Harvey is a freelance writer based in Wilmington, Delaware.