Scientists are developing a technology for diagnosing cancer pathologies at an early stage and without damaging the skin. The new approach makes it possible to detect biological tissue abnormalities characteristic of cancerous growths using special images obtained with a light scanner. For the patient, the procedure will resemble an ultrasound scan. It will not be a doctor who will be able to analyze the images, but artificial intelligence: this will minimize human error in the diagnosis. So far, the technology will not allow histological images of internal organs to be obtained, experts said. But the development can be used to diagnose, for example, skin cancer.
Physicians-histologists of the First Moscow Medical University. THEM. Sechenov, together with physicists from Great Britain (Aston University) and Finland (Oulu University), are developing a new way of digital diagnostics of neoplasms in humans.
Now with suspected oncology the surgeon usually takes a tissue sample from the patient at the site where the suspected tumor may be… Then a histologist works with the analysis, who makes the final diagnosis.
Scientists create digital histology technology that does not require biological tissue sections the patient. According to their idea, using a polarizing light scanner, it will be possible to obtain an image of a tissue site and detect oncology without making a section… Procedure resembles an ultrasound and will be painless for the patient. This technology is called histophotonics.
The specialist examines the glass with a stained histological section
Photo: RIA Novosti / Mikhail Kireev
Its use can also simplify the work of the doctors themselves.… Today, when analyzing a tumor, specialists use a microscope to analyze images of the thinnest sections of tissue taken from a patient.
“It takes a specialist three to four days to prepare a tissue sample (it is also called a drug) for analysis,” said a senior lecturer at the Department of Histology, Cytology and Embryology at I.M. THEM. Sechenov Gennady Piavchenko. – It consists of fixing, wiring, pouring, cutting, painting and drying cut glass. After the technician takes the microscope image, it must be reviewed by a practitioner.
At the same time, a physician can make a mistake when analyzing samples: the human factor cannot be excluded, the scientist noted. The oversight can be fatal for the patient.
Only light and nothing more
Scientists intend to gradually “digitize” the histology procedure stage by stage… Researchers are now studying the interaction of polarized light with diseased and healthy tissues, and are also developing a method for analyzing the resulting images using machine learning algorithms. The polarized light is separated from the general beam in such a way that the oscillations of the electric field in it occur only in one plane.
Photo: TASS / Valery Matytsin
“When light interacts with elements of biological tissue, the properties of polarization change differently for healthy and cancerous samples,” explained Igor Meglinsky, professor at NRNU MEPhI and Aston University. – Thus, by the changes in the characteristics of polarized light, one can judge the presence of structural abnormalities inside cells and the interaction between them. The most difficult thing is to identify and quantify the regularity of changes in the characteristics of polarized light during its interaction with biological tissues at the cellular level, which can be used in algorithms based on artificial intelligence and neural networks. We were able to do it.
So far, scientists have managed to track changes in the characteristics of polarized light in biological tissue by only a few millimeters.… I.e at this stage, the scanner can only analyze the skin or organs and tissues to which the probe can penetrate… However, you can also use the technology to study samples taken in the traditional way.
– The technology of non-contact histology seems very interesting, – noted the pathologist of the N.N. N.N. Blokhin Nikolay Kozlov. “But some types of laboratory diagnostics require a piece of tumor tissue. In particular, it is necessary for the analysis of the protein composition and mutations in the tumor, which determines the prognosis of the disease and the sensitivity to drugs. However, at the first stage, when it is necessary to find out about the presence or absence of a tumor, the technology can potentially be useful. However, the low penetrating power of the method makes it inapplicable for scanning the lungs, bone tissue and other deep localizations where a tumor may develop.
Artificial Intelligence Diagnosis
The developers of the new technology are also creating an artificial intelligence-based algorithm that can make a preliminary diagnosis from the image.… For this, scientists train a neural network. The computer is “shown” many pictures of healthy tissues and those affected by the tumor. Subsequently, an image of a biological tissue sample can be loaded into a computer program, the algorithm will compare it with the examples available in the database and make a presumptive diagnosis. With each new iteration, the neural network is trained and increases the accuracy of predicting the presence or absence of any pathology in the image, Igor Meglinsky specified.
Photo: TASS / Valery Matytsin
Now the reliability of the results of the neural network exceeds 90%, scientists say.… By the time the technique is released into clinical practice, they plan to bring the indicator as close as possible to 100%. Scientists expect to implement the technology and begin clinical trials within the next three years.
So far, doctors are skeptical about the capabilities of neural networks.
“If we turn to the results of the work of world scientific teams developing methods for making a diagnosis using computer analysis, we will not find successful results,” Nikolai Kozlov noted. – But such work has been carried out for more than 10 years. It is known that an experienced pathologist, when making a diagnosis, is able to make about 2% of errors per year.
It is premature to say that a neural network will be able to diagnose with an accuracy of 99.9%he concluded.