With increasing demand for highly precise image-based biopsies, the role of radiologists will become less siloed and more integrated with other disciplines.
Today, we’re seeing advancements in cancer therapies such as immunotherapy and gene editing that were unimaginable just a few years ago. We have a much better molecular understanding of cancer and the ability to target particular mutations across diseases or track it at the cellular level to diagnose the patient. However, in order to take full advantage of these technology advances, we need to do a better job on the diagnostic side. Most importantly, we need to integrate all the relevant information on a timely basis to be able to provide a selection of treatments that is based on the individual patient’s particular characteristics, not the general category of disease. This is key to personalized medicine, but it’s important to set expectations on enabling that future.
Getting Relevant Data to Decision-makers
In the near term, the main objective is getting all the relevant information (clinical history, imaging data, histopathology, and molecular pathology data) to the fingertips of decision-makers. We need to cull all this information together, because right now, it’s burdensome to do that manually. For personalized medicine (which includes imaging data, histopathology data, molecular mutation data and patient clinical history) this could be tens of thousands of pieces of information that needs to be aggregated for simultaneous visualization-a daunting task that no human being is capable of doing.
Making Sense of all the Data
Once we are able to more effectively collect all this voluminous data, the next phase is one in which we have to make sense of all the information. The first step is to resolve data discordances. When looking at data for a patient with lung or prostate cancer, for example, what a radiologist sees from an image may be very different from what a pathologist interprets from the data in terms of the stage of the disease. This is important because the aggressiveness of the tumor impacts treatment for the cancer patient. We must be able to review all diagnoses concurrently-and resolve differences or discordances-before we can move to the next step in making sense of all the data, which is finding associations.
Here’s where imaging technology advancements enable us to do much more in terms of building informatics and artificial intelligence around the data, so that we can understand the associations between the variables. Once we are able to make associations more readily, we can very precisely decide on the right therapy (or clinical trial) to choose for that patient, and be very predictive of outcomes for that patient.
Understanding associations between variables will help us distinguish what is really significant and what is irrelevant or false correlations. Overcoming obstacles to diagnostic confidence is critical because inconclusive studies delay definitive diagnosis, which delays treatment. For cancer patients, delay of treatment is both impactful on the clinical outcome and the emotional strain for patients and families. To that end, having both the informatics that integrates all of the data, as well as the adaptive intelligence that will sort through all of that and make sense of all that data in context, is highly important. Adaptive intelligence combines artificial intelligence and other methods with the knowledge of the clinical, operational, or personal context in which they are used. This will enable us to apply personalized medicine on a smaller scale to a very small number of individuals that have the same genotypic expressive (mutations at a molecular level) and phenotypic description (symptoms) of a disease.
Making New Discoveries from Data
Long-term, we will be able to leverage AI through adaptive intelligence to actually make new discoveries from the data. While this aspect of personalized medicine is much harder to do, it’s also the most promising in terms of what it will enable for the future of oncology in particular, and diagnosis and treatment in general.
For example, histopathology has historically largely focused on the structure of cells. When a cell becomes cancerous, its nucleus changes, its cellular membrane changes and its general characteristics change. So, when a pathologist looks at a slide, that’s what they look for- changes. However, it’s also quite clear that the changes don’t occur in isolation. The changes occur in the micro-environment in which these cells reside. This is where new discoveries can be made from data about diseases.
Understanding Disease Data from a Cellular, Molecular and Genomic Level
What researchers are starting to discover is that for certain types of diseases, lung cancer for example, there are changes in the micro-environment that are very difficult for human eyes to discern from an image. However, technology advancements are enabling machines to pick them out and correlate those changes with the aggressiveness of the disease. Similarly, there is evidence that breast MRI image textures correlate with the type of underlying mutations. Essentially, this means that if these correlations are sufficiently strong, we may be able to use image characteristics as a proxy for mutation analysis, and potentially save a biopsy. This allows us to potentially use more non-invasive procedures to reach a confident diagnosis for cancer patients, which is much more desirable.
In the past, we’ve made these discoveries on the basis of molecular and cellular understanding or on the basis of tissue-level understanding from radiology. As things evolve in the future and we continue to look at changes not only in isolation, but in micro-environments as well, we will have many more opportunities to discover new associations among these diagnostic parameters. This will help us predict how responsive a patient would be to a particular therapy and what the outcome would be, which is central to advancing the notion of personalized medicine. With these new discoveries being uncovered, treatment is becoming more and more molecularly driven, rather than location driven.
Changing Role of Imaging and Radiologists
As these molecular understanding and techniques get more advanced, the role of imaging will also change. Increasingly, there will be a convergence between the diagnostic understanding of disease, the cellular understanding of the disease (histopathology) and the molecular understanding of the disease (genomics). These disciplines are going to get closer and closer together and the interplay among these three key areas will get more important over the next five years.
As a result, the role of the radiologist will also change. If we are classifying more diseases by the molecular understanding of the disease, and that’s the primary determinant for which therapy to choose for the patient, then radiologists will have a more direct role in treatment and intervention. With increasing demand for highly precise image-based biopsies, the role of radiologists will become less siloed and more integrated with other diagnostic modalities.
Advancing the Future of Oncology, Diagnosis and Treatment
In order for the future of oncology to advance, we must find ways to better harness the tremendous potential of the increasing volume of information we’re collecting in order to drive better diagnosis and treatment, and ultimately, better patient outcomes. With this, the role of diagnostics will surely increase over time, because the cost of treatment is driving the cost of healthcare. We have no choice.
In order to make that happen, we need to not only aggregate more data, but also remove obstacles to aggregating that data. We need better disease models, and that really needs to be a collaboration between academia and industry in order to accurately assess disease progression and how it alters the diagnostic modality tests that we see. Lastly, the regulatory approval process really needs to be accelerated so that adaptive-intelligence-based algorithms that help us make sense of the massive amount of information and render a diagnosis can gain approval quickly. There needs to be incentives from the regulatory and reimbursement perspective to help healthcare providers adopt new business models that would show the value of AI technology adoption. These types of steps will help speed the delivery of personalized medicine and truly impact the future of oncology and diagnosis and treatment in a profound way.
Homer Pien, Ph.D., is the chief scientific officer, diagnosis and treatment, for Philips.