Advancing Imaging Research: Focus on Precision Medicine and AI
Precision medicine is advancing rapidly, in almost every field of medicine, and the idea that clinicians can—and should—tailor treatments to match each individual patient is driving ongoing large-scale clinical trials and many future studies.
But to fully achieve the promise of precision medicine, researchers will need to harness the power of “big data”—massive multi-modality biomedical datasets that are growing exponentially. Artificial intelligence (AI) coupled with high performance computing will be key to unlocking big data.
Big data, however, comes with a lot of small problems, namely the many logistical, technical and other issues arising when researchers set out to collect, store, clean, process, and explore large, multimodal datasets from multiple sources.
Recent reports indicate that at least 80% of effort in data science is linked to data preparation, leaving minimal time for research and analysis.
To accelerate the development of precision medicine, researchers need solutions to reduce the time they spend on managing data and IT tasks so they can focus on the research and analysis that are at the core of advancing precision medicine.
What is Precision Medicine?
Precision medicine—sometimes also called personalized medicine—is the ability to determine biological differences in diseases and the different effects they have in individuals, and then making treatment decisions based on that knowledge.
To understand how precision medicine differs from current medical practice, it can be useful to compare it to the current “trial-and-error” approach when clinicians choose a drug to treat a particular disease.
A doctor may prescribe any number of topical skin medications to treat psoriasis—but that decision is usually based on very little specific information about a patient’s “omics”—an understanding of the patient’s biology at different levels, including the molecular gene level (“genomics”), the protein level (“proteomics”), and the metabolic level (“metabolomics”).
If the prescribed medication doesn’t work in a few weeks, the clinician may prescribe another medicine, or supplement the first drug with another treatment.
In contrast, precision medicine has the goal of basing clinical decision-making on each patient’s unique biological makeup as determined through genetic tests or biomarkers. In our psoriasis example, molecular and cellular patterns, or phenotypes, could be used to learn which inflammatory cell phenotypes are driving a patient’s disease, permitting a physician to tailor a first-line treatment based on that information. Such strategies have the potential to get patients on the right medication sooner.
Oncologists have been on the forefront of adopting precision medicine models, in large part due to the ability of next-generation sequencing technology to rapidly and accurately sequence many genes at once.
Oncologists can now routinely sequence a patient’s tumor, and match the results with a chemotherapy or radiation therapy regimen designed to target the genetic alterations driving the tumor’s growth.
Immunologists also frequently link disease progression to the results of tests for patient-specific biomarkers, genotypes, and molecular profiles.
Big data and AI will drive advances in precision medicine
Precision medicine can therefore only become fully realized in clinical settings when researchers collect large multimodal biological datasets to identify genetic and other biological variations in patients and then conduct large-scale, long-term clinical trials to find treatment options with the best outcomes for particular subsets of patients.
Employing AI technologies is the perfect tool to accelerate that process of discovery. AI can now take advantage of high-performance computing capabilities, unavailable even a few years ago, and has led to the development of algorithms that can help predict risk and direct evidence-based treatment strategies for many medical conditions.
Advancing precision medicine through imaging research
Imaging science is a powerful method to get a high-level view of a patient or disease, and is currently underutilized in precision medicine research. Advanced imaging scans (CT, MR, ultrasound) harbor a wealth of data that until now has been largely unexplored. That is likely to change in the future as imaging -omics or radiomics contribute to our understanding of disease processes and the identification of the deep phenotypes that will make precision medicine possible. Advances that are already underway in molecular imaging modalities, which are rich in clinically relevant biomarkers, will further accelerate the field.
These molecular imaging techniques visualize biological processes in the body in real time, at the molecular level, and when combined with injected contrast agents can measure metabolism, oxygen status, blood flow, tumor cell growth, and even gene expression.
The high sensitivity and precision of molecular imaging can identify those biomarkers indicating pre-disease states or molecular states that occur before typical symptoms of a disease are detected. AI and machine learning are the right tools to discover which biomarkers can be used to deliver more precise, individual patient-focused treatments and better outcomes.
Challenges associated with big data and AI
Big data and AI will drive progress in precision medicine, but their success in creating true translational research and clinical advances is dependent upon addressing critical research data management needs. Machine learning requires large volumes of data for accuracy in most applications.
Institutions often have a wealth of data but lack the systems needed to get it into the hands of researchers cost-effectively. This hinders the development of machine-learning data sets, which gain statistical power as they grow. But if AI developers can gain access to imaging datasets created from multiple collaborating institutions, they can more quickly train their systems at the necessary scale.
There are six critical research data needs to accelerate AI development for precision medicine:
- Multimodality. Researchers will require a 360-degree view of patients including medical imaging, EMR, digital pathology, EEG, -omics, and other data. Systems to collect clinical datasets used in precision medicine-focused research must deal equally well with clinical imaging and with nonimaging data, particularly for research workflows.
- Shared projects and cohorts. Researchers must have the ability to organize and analyze data in cohorts while enabling collaboration with others outside of the context of clinical care. However, clinical data systems are often designed for individual patient care, not for cohort or population health studies.
- Quality assurance. Machine learning and AI researchers must be able to confirm the quality of data, including completeness and consistency with the protocol defined for the study. Because quality control and supporting documentation are required for scientific reproducibility and for regulatory approval, systems supporting research on large data sets must have comprehensive support for provenance as part of the workflow.
- Integrated metadata, labeling and annotation workflows. Machine learning depends on accurately labeled sample datasets in order to effectively train AI models. Researchers must have the ability to efficiently organize and normalize classification for search and selection into the appropriate collections for machine learning training.
- Automated computational workflows. Machine learning and AI algorithms are computationally intensive activities, and so researchers must be able to automate and scale computational workflows for analytic pipelines and AI model training. This can be done using cloud-based or high-performance computing infrastructures to process large cohorts cost-effectively.
- Integrated data privacy and IRB compliance. Patient data may lack consent for use in research. Researchers must therefore have systems in place to enforce IRB constraints and regulatory compliance. Researchers must also have access to systems that address data privacy by de-identifying protected health information at a level required for each project.
A case study on accelerating precision medicine with imaging research
One study that has already produced clinically relevant results examined images from more than 3,000 patients with traumatic brain injury (TBI), gathered from 19 participating institutions.
With the goal of developing a precision medicine model to improve the diagnosis, treatment, and rehabilitation of patients with TBI, Dr. Geoffrey Manley, Vice Chairman of Neurological Surgery at the University of California, San Francisco, and colleagues set up the Transforming Research and Clinical Knowledge in Traumatic Brain Injury (TRACK-TBI) study.
Dr. Manley and colleagues used Flywheel as a cloud-based centralized platform that aggregates and securely shares medical imaging and related clinical data across multiple sites.
The TRACK-TBI study has found indications that a technique called neurite orientation dispersion and density imaging (NODDI) may be a more sensitive biomarker for mild TBI than previous methods.
“With our imaging data managed on Flywheel, we have explored diagnostic models grouping biomarkers discovered in the TRACK-TBI study and hope to one day see more individualized treatments for brain injury,” says Dr. Manley.
Access a webinar on the TRACK-TBI study presented by Dr. Manley and Sina Aslan, Ph.D., Scientific Advisor at Flywheel.
The way forward to precision medicine
The promise of precision medicine is to use large imaging-based and multi-modal datasets to discover individual variability and then analyze those datasets to develop treatments tailored to each individual patient.
Key to reaching the full potential of precision medicine is taking advantage of machine learning and AI algorithms.
Doing so requires researchers to efficiently collect, store, clean, process, and interpret large, multimodal datasets from multiple sources, which come with many logistical, technical and other issues.
Research informatics platforms can help solve these challenges and improve collaboration with easier data sharing and reuse. Ultimately, research data platforms can contribute to faster discoveries in precision medicine, and better outcomes for patients.
Schedule a Demo
Flywheel improves research productivity, accelerates innovation, and helps researchers advance the science of precision medicine by curating AI-ready data, streamlining data analysis, automating data management and storage, and enabling data sharing and collaboration. Learn more with a custom demo.