Integrated diagnostics can provide a wide variety of potential benefits to patients facing cancer. It can also give radiologists the opportunity to become more involved in clinical decision support beyond imaging. Speakers addressed the current and future state of integrated diagnostics at Tuesday's RSNA/AAPM (American Association of Physicists in Medicine) symposium.
"The development of new and more sophisticated approaches to diagnostic testing, including medical imaging, anatomic pathology and laboratory medicine, along with the growth in targeted cancer therapies, is transforming the landscape of cancer diagnosis and care," said Paul E. Kinahan, PhD, symposium moderator, vice chair for radiology research and head of the imaging research laboratory at the University of Washington, Seattle.
Using Tools and Data to Aid Diagnosis
Radiology needs to be empowered to embrace integrated diagnostics through tools, standards and data science, according to Mitchell D. Schnall, MD, who presented The Path to Integrated Diagnostics.
"Developments in artificial intelligence and data science have created additional opportunities to extract and combine information from imaging, pathology, laboratory medicine and genomics to create an opportunity to improve diagnosis by providing access to a broad spectrum of information on each patient," said Dr. Schnall, the Eugene Pendergrass Professor of Radiology at the Perelman School of Medicine at the University of Pennsylvania, Philadelphia. "As a result, radiologists have access to more non-imaging data than we have ever had before. Integrated diagnostics is slowly becoming the norm of radiology practice, as we integrate laboratory data with our imaging data and clinical findings in our daily work."
Dr. Schnall acknowledged that current approaches to finding data in medical records are cumbersome and time consuming. While there are few standard approaches to integrating the information, the hope is that new standards will evolve to help radiology assist in improving diagnostics.
Cross-Disciplinary Implementation of Integrated Diagnostics
Regardless of choice of AI approach — radiomics or deep learning — there is a need to establish multi-scale disease associations, said Anant Madabhushi, PhD, in his lecture, Radio-Patho-Genomics: Computationally Integrated Disease Specific Features Across Scales.
"We need to establish high fidelity ground truth for disease extent on the radiographic imaging to be able to train AI models. However granular, 'ground truth' definition of disease extent is only available on surgical pathology specimens. AI can help in co-registering the ex-vivo specimens with pre-operative imaging, such as rectal, prostate, lung or breast," said Dr. Madabhushi, the F. Alex Nason Professor II of Biomedical Engineering at Case Western Reserve University and a research scientist at the Louis Stokes Cleveland Veterans Administration Medical Center, both in Cleveland.
AI can also be used to create predictors for identifying patients who are likely to have disease recurrence, progression or metastasis, according to Dr. Madabhushi, who has used a combination of AI extracted features from pathology images and CT scans to allow for better recurrence prediction of early stage lung cancer.
The ability to link and associate genomic and molecular information with radiographic patterns on imaging is a future benefit of integrated diagnostics.
"AI can help in identifying association between genomics, pathology and imaging features," Dr. Madabhushi said. "For instance, there is a great deal of interest currently on developing AI for radiogenomics and using AI to help identify associations between imaging features and point mutations and biological pathways."