The Congressionally Directed Medical Research Programs funds research to modernize cancer detection and diagnosis methods to prevent lethality of cancer in Service Members and Veterans. As part of their military duties, Service Members may encounter environmental or occupational exposures associated with increased cancer risk, or cancer may develop without a known cause during or after service. Regardless of cause, cancers that return after initial treatment or remain undetected for an extended period may metastasize, or spread, to other parts of the body, impacting survival.
Through partnerships with the U.S. Department of Veterans Affairs and data collected from Veterans, CDMRP-funded researchers advance artificial intelligence for early detection or prediction of cancer spread or return, which can inform clinical decision-making to reduce progression, prevent recurrence and increase survival in Service Member and Veteran populations. Further, thanks to the structured nature of health care provided to Service Members and Veterans, samples from these patients may include well characterized histories from service records and electronic medical records, increasing the ability to identify patient-specific associations with potential cancer risk.
When suspecting cancer during a patient examination, clinicians order imaging tests to help identify the location of tumors and tissue biopsy to collect cells from concerning regions for closer examination. Irregular patterns in tissue shape, thickness, texture and structure inform initial cancer diagnosis and staging. Artificial intelligence and machine learning algorithms offer opportunities to improve detection of these features, and even to identify patterns not seen before.
Visual Analysis of Prostate Cancer Cells for Rapid Diagnosis and Staging
The VA supports a biorepository of clinical and digitized imaging data and various patient samples from Veterans with prostate cancer or at risk for developing the disease. Approved researchers can access the biorepository, called VA-MAPP for VA-Multi-OMICS Analysis Platform for Prostate Cancer, through a secure computing environment, allowing for data analysis of thousands of prostate cancer cases from Veterans. In fiscal year 2020, the Prostate Cancer Research Program funded a Population Science and Outcomes Research Award, led by Beatrice Knudsen, M.D., Ph.D., and her team at the University of Utah, to develop a machine learning algorithm to help pathologists detect and stage prostate cancer from prostate tissue samples stored in VA-MAPP. Pathologists usually diagnose prostate cancer by examining prostate tissue biopsy samples or surgical specimens under a microscope.
The algorithm, called https://www.sciencedirect.com/science/article/pii/S0893395224000279?via%3Dihub, mimics the pathologist’s manual workflow in three steps. HistoEM first differentiates the prostate gland from surrounding tissue, then classifies regions of the tissue as healthy or cancerous based on specific tissue and cellular features, such as shape and thickness. Finally, for cancerous regions, the algorithm determines whether the cancer is low or high risk of metastasis based on cancer staging criteria and complexity of the gland cell and tissue structures.
Importantly, HistoEM detects tissue patterns across the entire prostate gland, including concerning areas requiring closer examination by a pathologist, providing a complete picture to aid in diagnosis and treatment planning.
In addition to high or low risk, some prostate cancers fall into an intermediate risk category, complicating treatment planning. Non-cancerous cells near prostate tumors undergo genetic and structural changes to support cancer growth, https://www.mdpi.com/2072-6694/16/21/3685 disease progression and metastasis. Evaluating biomarkers within the tumor environment together with tumor-based markers may improve outcome prediction and treatment planning. In fiscal year 2024, the Prostate Cancer Research Program funded a Data Science Award – Partnering PI Option, led by Massimo Loda, M.D., at Cornell University, and Edward Giovannucci, M.D., at Harvard University, to develop AI algorithms for identifying cellular changes in the tumor environment that could indicate risk of progression to lethal prostate cancer.
First, the team will evaluate samples from the VA-MAPP with genetic mutations known to increase risk for metastasis and use machine learning to look for visual patterns unique to the mutated cells. Then, they will develop and test an algorithm to see how well the patterns correlate with, and potentially predict, disease progression including metastasis.
If successful, this visual analysis of cells in the tumor environment could bypass the need for extensive genetic analysis in diagnosis and staging of intermediate prostate cancer, allowing for more rapid determination and initiation of appropriate treatment plans.
Image-Based Detection of Population-Specific Risk in Head and Neck Cancer Recurrence
About 80 percent of people living with oropharyngeal cancer, a cancer occurring in the back of the throat, tongue or tonsils, receive their diagnoses after tumor cells metastasize to nearby or distant parts of the body, significantly impacting response to treatment and patient outcomes. Survival outcomes also vary across different populations, including Veterans.
In fiscal year 2020, the Peer Reviewed Cancer Research Program funded a Career Development Award, led by Germán Corredor, Ph.D., at Emory University, to develop DigiTIL, an AI tool for analyzing tumor cells and other cells within the tumor environment to calculate a recurrence risk score for each patient sample. Corredor and his team, including partners at the Houston VA Medical Center and Louis Stokes Cleveland VA Medical Center, built DigiTIL using tissue images from over 1,000 well-characterized oropharyngeal cancer cases, including metastatic cases, collected from six institutions and two clinical trials. When comparing the algorithm’s calculations to known patient histories, DigiTIL accurately assigned high risk scores to oropharyngeal cancers that recurred after initial treatment, and low risk scores to cases that responded well to standard of care therapy, demonstrating the tool’s predictive ability to specifically identify cancers likely to return.
DigiTIL also https://acsjournals.onlinelibrary.wiley.com/doi/10.1002/cncr.34446 population-specific tissue biomarkers associated with poor disease outcomes, and in preliminary results DigiTIL identified significant differences in cancer tissues between Veteran and non-Veteran samples.
Image-based diagnostic tests like DigiTIL could help clinicians incorporate individual and population-based risk factors to determine patients' risk of cancer recurrence and inform treatment planning.
Lung Tumor Imaging Informs Routine Surveillance for Brain Metastasis
Lung cancer presents with a high incidence of brain metastasis. Patients with stage I and II non-small cell lung cancer typically do not undergo brain imaging, such as MRI, at diagnosis and rarely receive yearly surveillance brain scans, making new and spontaneous brain metastases challenging to detect.
In fiscal year 2021, the Lung Cancer Research Program funded an Investigator-Initiated Translational Research Award, led by Laura Stabile, Ph.D., at the University of Pittsburgh, to explore whether radiogenomics, a technique that integrates cancer imaging with genomics data, holds information to predict lung cancer patients likely to develop brain metastases months to years after initial diagnosis.
Stabile and her team, including partners at the VA Pittsburgh Health System’s Lung Cancer Precision Oncology Program, https://virtual.oxfordabstracts.com/event/74542/poster-gallery/grid?fullScreen=false¤t=120 different machine learning models for predicting brain spread based on computed tomography, or CT, radiogenomic analysis of lung tumors from 484 patients, half of which showed brain metastases.
The best-performing model reached about 80 percent accuracy for predicting lung tumors that spread to the brain, and worked especially well for early-stage patients, showing who might later develop brain metastases. The model also distinguished brain-specific metastatic potential from general cancer spread. For patients with current brain metastases, the model accurately identified the presence of brain tumors from the initial CT radiogenomic analysis of the lung tumor.
If validated in larger groups, this tool could help doctors decide which patients need more frequent brain MRIs and who would benefit from earlier treatment to prevent brain metastasis. For more information about the Prostate Cancer Research Program, visit https://cdmrp.health.mil/pcrp/.
For more information about the Peer Reviewed Cancer Research Program and the program’s 22 congressionally directed topic areas in fiscal year 2025, including lung cancer, visit https://cdmrp.health.mil/prcrp/.
For more information about the Lung Cancer Research Program, funded between fiscal years 2009 and 2024, visit https://cdmrp.health.mil/lcrp/.
| Date Taken: | 12.02.2025 |
| Date Posted: | 12.02.2025 14:00 |
| Story ID: | 552754 |
| Location: | US |
| Web Views: | 19 |
| Downloads: | 0 |
This work, CDMRP Advances AI-Based Metastatic Cancer Detection Through VA Partnerships, must comply with the restrictions shown on https://www.dvidshub.net/about/copyright.
