June 1, 2023 | Using a special tissue grinder, label-free cytometry, and machine learning, researchers in Germany have come up with a way to evaluate potentially cancerous tissue without a trained human pathologist in only 30 minutes—fast enough to be done during surgery. The unconventional method looks at single cells rather than the entire architecture of tissue to decide whether an organ is diseased or not, according to Despina Soteriou, Ph.D., a lab scientist at the Max Planck Institute for the Science of Light (MPL) in Erlangen.
The feat was demonstrated in a study, conducted jointly with her colleague Markéta Kubánková, Ph.D., which published recently in Nature Biomedical Engineering (DOI: 10.1038/s41551-023-01015-3). A microfluidic technique known as “real-time deformability cytometry” (RT-DC), developed in the lab of MPL Director Prof. Jochen Guck, was used for the cell imaging step.
The vision here is to fashion a device to help surgeons in the operating room know if tissue is cancerous and, if so, whether they have reached the margins of the healthy part of the organ, says Soteriou. In the future, the same “artificial pathologist” might also perform a more detailed examination of the tissue to determine the exact type of cancer the patient has to further guide treatment planning.
Currently, surgeons rely on an on-site pathologist to assess solid tissue biopsy samples, she continues. But pathologists are in short supply in many parts of the world (including Germany and the United States) and, unlike a programmable machine, don’t work around the clock or without pay.
Automated tissue analysis is enabled by a trifecta of innovations, starting with the grinder developed at the Fraunhofer Institute for Process Automation (IPA) in Mannheim. Soteriou says it operates on the same principle as a wheat grinder, with a speed adjuster, for quickly tearing down biopsy samples to the single-cell level without damaging the cells.
It’s the necessary first step before moving on to RT-DC, which is similar to flow cytometry in that single cells in suspension are being analyzed except images are being taken of each one—up to 1,000 per second, 36,000 times faster than traditional methods for analyzing cell deformability, she explains. The cells first pass at high speed through microfluidic channel constriction, where they deform by shear stress and pressure gradients.
Much information about the physical parameters of the cells can be derived from the images produced, including how large and deformable they are, explains Soteriou. These properties may change during disease. “In patients we first started looking at blood because cells are already suspended in liquid and easier to analyze than cells tightly bound in a piece of solid tissue.
“From these first studies it became clear that the detection of different disease states is possible,” Soteriou continues. “This is why we came up with the idea of [using RT-DC]... to look at cells from solid tissues.”
The researchers started their study on mouse samples. Encouraged by the results, they analyzed biopsy samples from over 30 colon cancer patients using a machine learning approach to analyze the output.
“For this paper, we wanted to look at how all parameters together can affect the phenotype of the disease state,” Soteriou says. Cell deformability was affirmed as an important biomarker in the mix. The combination with other parameters allowed the algorithm to reliably distinguish samples containing tumor cells from healthy tissue.
In the future, when researchers hope to repeat the study on a larger cohort of patients, the plan is to use artificial intelligence to extract more information from the data, she says. This could include the cancer stage, the cancer’s aggressiveness (or metastatic potential), or the survival rate.
Based on its performance in the latest study, the artificial pathologist could one day be employed in the absence of a human pathologist, says Soteriou. Doctors always need to coordinate surgeries based on the availability of a pathologist, which is limited. As imagined, the new diagnostic method could be performed by a nurse when a surgery needs to be scheduled and the pathologist has a conflict.
The method might additionally help clinicians assess disease severity or recognize different types of inflammatory bowel disease (IBD). Researchers used their technique to analyze colon tissue in mouse models of IBD and found that cell deformability reflects the severity of inflammation, Soteriou says.
A growing body of evidence suggests chronic inflammation is associated with malignancy and they therefore correctly speculated that the approach might also detect changes in biopsy samples from tumors. This possibility was confirmed in both mouse and human samples.
The plan now is to continue working closely with the clinicians at the University Hospital Erlangen to directly compare the performance of the human and artificial pathologist side by side in terms of speed and accuracy, says Soteriou. Challenges on the human side include time pressures and the number of tissue slices that can reasonably be examined under a microscope.
Further proof that the method overcomes these issues should facilitate its movement forward to the clinic, she says. While it might take a little work to convince pathologists to “try something different,” doctors are quite excited.
The biggest technical challenge will be integrating the tissue-processing and single-cell-phenotype analysis into a single automated pipeline, Soteriou adds. Combining the two devices into a single unit will require a lot of teamwork with the IPA.
As the Max Planck Society is a non-profit organization focusing on fundamental research, the company Rivercyte was recently founded to commercialize deformability cytometry in the medical field. It is working on turning the research device into an in vitro diagnostic medical product.