Talk to the Veterans Crisis Line now
U.S. flag
An official website of the United States government
Veterans Benefits and Health Care About VA Find a VA Location

Office of Research & Development

 print icon sign up for VA Research updates

Computer Vision and Machine Learning in Precision Oncology (CoMPL)

Computer Vision and Machine Learning in Precision Oncology (CoMPL) is a VA hub established in 2021 with the aim of improving clinical decision making for cancer and other diseases facing veterans using advanced machine learning algorithms.

CoMPL's Objectives:

  1. Increase the computational infrastructure and access to tools for computer vision and machine learning (CVML) within the VA.
  2. Build a community to enable VA researchers to take advantage of these tools to develop their own CVML applications.
  3. Develop new companion diagnostic tools for risk assessment and development of treatment plans.

Precision Oncology and How Computer Algorithms Can Help

The goal of precision oncology is to create targeted therapies specific to each patient’s needs. CoMPLs partner precision oncology initiatives within the VA include the Lung Precision Oncology Program (LPOP) and the Precision Oncology for Cancer of the Prostate (POPCAP) programs.

Computer vision (CV) refers to the ability of computers to extract information from digital images. In the clinical setting, these images include radiology images (MRI, CT, etc.) and digital pathology images (high resolution scans of microscope slides containing biopsy and other tissues).

Machine learning (ML) is a field of computer science in which computers are trained to see new patterns in data using advanced networks.

CVML can aid clinicians with precision oncology in the following ways:

  • Reading patient images: Patient images include those obtained from MRI and CT scans as well as pictures taken of biopsy tissue. After images are collected, radiologists and pathologists examine them and inform the clinician about what the images show. These specialists may look at hundreds of images a day, and some features of these images may be very difficult or impossible to evaluate by eye. Computers using machine learning algorithms may aid scientists with tasks such as counting cell types in a biopsy image or by identifying missed or new features of the images that are not easily detectable by the human eye.
  • Combining genetic and molecular data: Clinical tests such as blood draws and biopsies provide information about what genes are active in the patient. Results of these tests may help clinicians make predictions about the Veteran's prognosis or potential response to treatment. Machine learning algorithms may be used to assist clinicians in determining new patterns from test results.
  • Creating super-classifiers: In addition to radiology and pathology images, factors such as age, race, and other social determinants of health can affect patient prognosis and response to treatment. Scientists can use machine learning algorithms to create super-classifiers based on these factors. Clinicians may than use these classifiers to aid in the creation of personalized treatment plans for each patient based on their individual risk profile.

Examples of Computer-Assisted Diagnosis and Detection

Radiology Images

MRI and CT scans generate a series of two-dimensional images of internal anatomy that can be used to create a three-dimensional picture. Computer vision can use either the 2D or 3D images to determine which features are important for disease prognosis, progression, or response to treatment. For cancer patient data, some features that can be found from these images are the size, shape, and texture of the tumor, as well as whether it has spread to other regions.

The environment of the tumor is also examined to help understand if and how the tumor will spread. For example, it is known that cancer induces the cardiovascular system into re-directing blood vessels to feed the tumor. Computer algorithms can be used to detect and quantify detailed features of the blood vessels that correlate with cancer prognosis.

At left is a 3-dimensional reconstruction of lung blood vessels in a cancer patient generated from CT scans. Researchers examined the shape and size of the blood vessels, as well as how often they twisted and turned as they reached the tumor. At right, blood vessels surrounding the tumor are shown from two lung cancer patients, one of whom did not have the cancer return after treatment (left) and one who did (right). Researchers found that patients who did not have the cancer return had fewer twisted blood vessels.

At left is a 3-dimensional reconstruction of lung blood vessels in a cancer patient generated from CT scans. Researchers examined the shape and size of the blood vessels, as well as how often they twisted and turned as they reached the tumor. At right, blood vessels surrounding the tumor are shown from two lung cancer patients, one of whom did not have the cancer return after treatment (left) and one who did (right). Researchers found that patients who did not have the cancer return had fewer twisted blood vessels.1

Pathology Images

A biopsy is a sample of patient tissue that can be placed on a microscope slide and stained so it can be examined by a pathologist. In digital pathology, biopsy slides are scanned by a high-resolution scanner so that they can be viewed at high magnification by pathologists and computers without a microscope. Features of the tissue that are relevant to prognosis include the general organization of the tissue sample, the size and shapes of cells, signatures of the cells that show how fast the cells are dividing (mitosis), and the structure and orientation of the fibers that hold the tissue together (collagen protein fibers).

Computers can be used to quantify certain features of the tissue that might be difficult for a pathologist to describe. Some of these features have clinical relevance. For example, the level of order of collagen and the number and distribution of immune cells in the tumor have both been shown to be associated with cancer recurrence and disease-free survival after treatment. This is because tumor cells can travel along ordered collagen fibers in the tissue, while cancer-fighting immune cells have more luck getting into the tumor when the fibers are less ordered.

Left image: The image at left shows examples of collagen fiber arrangement in two breast cancer tumors. The top set of images show tissue from a patient with a short disease-free survival after treatment; collagen fibers (highlighted with blue lines) are much more orderly. The bottom image shows a patient with long disease-free survival. The collagen fibers in this patient were much more disordered. The lack of order meant this tissue had a less favorable environment for cancer progression. Top image: The tissue sample shown above was taken from a lung cancer patient who had cancer that did not return after treatment. A computer algorithm was applied to the original image of the tissue (right) to detect different kinds of cells (left). In the left image, cancer cells are shown in cyan and cancer-fighting immune cells called lymphocytes are shown in yellow. The algorithm showed that patients with lymphocytes evenly spread throughout their tissue were much likely to have their cancer return after treatment than those whose lymphocytes were only found in clusters.

Diseases Currently Under Investigation

  • Lung Cancer
  • Lung Diseases
  • Prostate Cancer
  • Head and Neck Cancer
  • Gastrointestinal Cancers
  • Cardiovascular Disease

Data Collection and Computing

CoMPL Research in the News

CoMPL Contact Information

Director: Anant Madabhushi, PhD
Administrative Director: Michael Gilkey, MS MBA
CoMPL@va.gov

For more about the VA’s commitment to ensure the deployment of safe and secure AI in VA healthcare, see the following: VHA Aligns with Leading Health Care Organizations to Ensure Trustworthy Use of AI.

References:

1. Braman N, Prasanna P, Bera K, Alilou M, Khorrami M, Leo P, Etesami M, Vulchi M, Turk P, Gupta A, Jain P, Fu P, Pennell N, Velcheti V, Abraham J, Plecha D, Madabhushi A. Novel Radiomic Measurements of Tumor-Associated Vasculature Morphology on Clinical Imaging as a Biomarker of Treatment Response in Multiple Cancers. Clin Cancer Res. 2022 Oct 14;28(20):4410-4424. doi: 10.1158/1078-0432.CCR-21-4148. PMID: 35727603; PMCID: PMC9588630.

2. Li H, Bera K, Toro P, Fu P, Zhang Z, Lu C, Feldman M, Ganesan S, Goldstein LJ, Davidson NE, Glasgow A, Harbhajanka A, Gilmore H, Madabhushi A. Collagen fiber orientation disorder from H&E images is prognostic for early-stage breast cancer: clinical trial validation. NPJ Breast Cancer. 2021 Aug 6;7(1):104. doi: 10.1038/s41523-021-00310-z. PMID: 34362928; PMCID: PMC8346522.

3. Corredor G, Wang X, Zhou Y, Lu C, Fu P, Syrigos K, Rimm DL, Yang M, Romero E, Schalper KA, Velcheti V, Madabhushi A. Spatial Architecture and Arrangement of Tumor-Infiltrating Lymphocytes for Predicting Likelihood of Recurrence in Early-Stage Non-Small Cell Lung Cancer. Clin Cancer Res. 2019 Mar 1;25(5):1526-1534. doi: 10.1158/1078-0432.CCR-18-2013. Epub 2018 Sep 10. PMID: 30201760; PMCID: PMC6397708.

top



Questions about the R&D website? Email the Web Team

Any health information on this website is strictly for informational purposes and is not intended as medical advice. It should not be used to diagnose or treat any condition.