Allina Health Cancer Institute and Astrin Biosciences are jointly integrating AI to improve the assessment of cancer cells in blood samples, beginning with breast cancer, which faces challenges with mammogram imaging and demands innovative solutions.
Artificial intelligence (AI) serves as a valuable tool in analyzing a standard blood sample to detect cancer cells, helping determine the specific cancer type or predict potential development of a present cancer. Collaboratively, Allina Health Cancer Institute in Minneapolis, Minnesota, and Astrin Biosciences are actively integrating AI to enhance the assessment of cancer cell presence within blood samples, starting with breast cancer, which faces a unique issue with the shortfalls of mammogram imaging and requires a transformative approach.
“What the human brain and eye are doing is using any clinician’s repertoire or stored database of images when looking at an x-ray or blood sample, but this [method] is limited because it’s whatever that clinician has seen in their lifetime or in their training. AI can be programmed to utilize multiple repertoires as well as continue learning,” Badrinath Konety, MD, explained in an interview with Targeted Therapies in Oncology. Konety is president of Allina Health Cancer Institute.
At Allina Health Cancer Institute, blood samples are drawn into standard blood tubes and then sent to Astrin’s laboratories. The samples are then processed through proprietary optofluidic chips, where the sample traversing through microcapillaries is subject to high resolution laser microscopy,” Konety explained. A holographic 3-dimensional image of each cell is generated, and “this is where AI comes into play,” Konety said.
The AI program evaluates the holographic 3-dimensional image of the cell, comparing it with a stored database of scanned proven cancer cell images Konety explained. Cells that have features consistent with a cancer cell are sorted out and collected. Simultaneously in the identification process, cells are also stained with different types of antibodies, some specific to a cancer type. “Not only is the AI evaluating and cross-referencing the shape, size, and nature of the cell, but it’s also determining whether the cell is expressing specific markers for a certain cancer type, and this adds a double layer of identification,” Konety noted. “If you use AI to find cells that are more specific [to the type of cancer you’re evaluating for], you can detect abnormalities early.”
This process is possible because of the computing power behind the approach and growing database of cancer cells, Konety explained. “We’ve trained the AI model on many known cancer cells so that it has this database to compare [with], and each time we run a new sample through, we input the sample criteria into the database.” As a result, this database grows, enabling increasingly comprehensive comparisons for each new sample.
“We’re doing something similar but using AI to scale up the level of comparison,” Konety noted. The method improves the fidelity of the traditional method of one pathologist looking at a slide and saying there is evidence of cancer. “Now you’re taking that same sample and evaluating the markers, genetic makeup, and the protein makeup and converging multiple techniques to improve the accuracy, and that enhances the specificity of the diagnosis,” Konety said. There is also a heightened sensitivity to this method, Konety explained. Because each cell in a sample of blood is evaluated in the form of a large 3-dimensional image, it’s a more thorough analysis and “similar to looking for a needle in a haystack; however, with this method, when you find the needle, you have a better way of knowing it’s a needle and not a pin,” Konety said.
Other medical companies are using AI to enhance traditional methods of analysis, including liquid biopsy. Epic Sciences has developed 3-dimensional imaging models to better evaluate the population of cells. RareCyte, Therapanacea, and Guardant Health are other companies making strides in the field to help diagnose patients.
A similar concept is an FDA-approved test, designed by CELLSEARCH, to identify circulating cancer cells found in the blood. However, the technique differs in that CELLSEARCH uses tubes that are coated with an antibody called epithelial cell adhesion molecule (EpCAM) to identify the epithelial cell. Konety explained that with this method the specifi city and sensitivity of the test are not as precise because the cells may or may not have that antibody or they may not have enough of the antibody to be captured.1
During the development and initiation phase of this method, Konety and Astrin Biosciences’ CEO, Jayant Parthasarathy, PhD, were looking for a way to take this approach to the bedside to help patients. Interestingly, breast cancer has a unique concern. Approximately 40% of women have portions of dense fibrous tissue within their breasts. This density prevents a clinician from discerning whether a tumor is present on a mammogram.2 The next step would be to order an MRI. The FDA has mandated that all facilities must update protocol to include a statement noting that the mammogram may not be accurate, and patients are recommended to undergo other testing to determine whether they have cancer present, Konety explained.3
In combination with this concern, Konety said that when evaluating breast cancer samples in the laboratory, “surprisingly, even in the early stage of breast cancer, we were able to fi nd cancer cells in the bloodstream, which we thought was unusual.” Konety and Parthasarathy researched further and found that others have reported on this phenomenon, finding circulating cancer cells early on in both bone marrow and blood.4 In efforts to confirm these findings, the duo started evaluating a larger subset of patients and processing isolated cells for gene expression signatures specific to breast cancer. “We were able to prove that these are indeed breast cancer cells because they express [with] positive [results] for all these markers,” Konety said.
Physicians managing patients with ductal carcinoma in situ (DCIS) face a conundrum because not all patients with this type of breast cancer will develop an aggressive cancer. “Many women are observed, and nothing happens to them,” Konety noted, and because there currently isn’t an ideal alternative solution apart from an MRI for women with dense tissue in their breasts, this simple blood evaluation acts as a second verification before an MRI.
“If we have a blood test that can identify the women who have the highest risk, then maybe this will make a difference,” Konety said. He surmised that women who have many or any cancer cells present in the blood may be those who experience disease progression and need an MRI and those who have few or no cancer cells present in the blood may not need an MRI because the cancer is less likely to progress.
The convergence of various concerns in breast cancer management presents an ideal opportunity to validate and define the role of AI in liquid biopsy. This technology stands to aid patients for whom mammograms may not be effective in identifying whether DCIS poses a risk.
Using AI, Konety noted, can also give oncologists a closer view of how the proteins react to cancer therapies and thereby refine treatment approaches for the individual patient. Allina Health Cancer Institute and Astrin Biosciences are set to report their findings later this year.
“In the future, there is potential for AI to replace traditional methods. However, for now, its role primarily involves augmenting our established approaches,” Konety said.