Russian Scientists Speed Up Melanoma Detection With AI
Researchers in Russia have developed a neural network that analyzes electron microscopy images of melanoma cells 15 times faster than manual analysis. Instead of taking 20–30 minutes to process a single image, the analysis now takes two minutes.
Melanoma is a malignant tumor that develops from the skin’s pigment-producing cells. It is considered one of the most aggressive forms of cancer because of its ability to metastasize rapidly. Microscopy is one of the primary methods used to diagnose it. But examining cancer cells under an electron microscope is painstaking and slow. A clinician may spend up to 30 minutes on a single image. And there can be hundreds of images, with a human life behind each one.
Researchers at the Institute of Cytology and Genetics of the Siberian Branch of the Russian Academy of Sciences and the Research Institute of Clinical and Experimental Lymphology developed a neural network that reduces processing time for a single image to two minutes. That makes the work 10–15 times faster. The results have already been published in the international Journal of Imaging.
How the Neural Network Was Developed
The developers trained the algorithm on hundreds of electron microscopy images of mouse melanoma cells. This type of tumor was chosen for a reason: it has long served as a standard model in laboratory experiments, giving scientists an extensive collection of examples for training the neural network.
The algorithm is based on computer vision techniques. The researchers were able to make the model perform consistently even though images produced by electron microscopes often contain defects and poorly defined boundaries. The algorithm now operates reliably under these conditions.
What Does the Algorithm See?
The researchers focused on two intracellular components: mitochondria, which produce energy, and the endoplasmic reticulum, which serves as a reservoir for calcium ions. Their interaction plays a critical role in cell signaling. When this connection is disrupted, tumor cells stop receiving signals that trigger programmed cell death, or apoptosis.
The contact sites between these structures are what scientists consider one of the most promising targets for anticancer therapy. Targeting them could help stop the uncontrolled proliferation of cancer cells and bring the disease into remission.
Melanoma and Beyond
For now, the Russian technology has been adapted exclusively for melanoma. But the researchers plan to expand its use to other tumor types and adapt the algorithms to analyze human biopsy material. That could pave the way for using the neural network not only in fundamental research but also in routine clinical practice. And it will certainly save more than a hundred lives.
The researchers have also built a web service around the technology that generates a table containing the analysis results. In the future, tools like this could become standard in research centers where much of this work is still performed manually. The team also plans to integrate such algorithms with digital pathology and medical imaging analysis systems.
Computer Vision in Microscopy
The Siberian researchers’ work is part of a broader push to digitize Russian medicine and, in particular, oncology services, one of the most important areas of healthcare worldwide.
In 2023, Russian scientists created a dataset of annotated microscopic images of cancer cells for training computer vision systems. At Sechenov University, researchers trained a neural network to detect colorectal cancer metastases in digital histology slides. In 2024, researchers at Southern Federal University developed a method for processing micrographs that cut the error rate in determining the number of neighboring cells in half.
In 2025, researchers at Peter the Great St. Petersburg Polytechnic University developed a neural network that automatically corrects optical distortions in microscopic images. The Siberian technology offers further evidence that artificial intelligence is on track to become an essential tool in laboratory diagnostics.
