AI detects cancer and viral infections with nanometer precision

Researchers have developed artificial intelligence that can differentiate between cancer cells and normal cells, as well as detect the very early stages of a viral infection inside cells. The findings, published today in a study in the journal Nature Artificial Intelligencepave the way for better diagnostic techniques and new disease monitoring strategies. The researchers come from the Center for Genomic Regulation (CRG), the University of the Basque Country (UPV/EHU), the International Physics Center of Donostia (DIPC) and the Biofisica Bizkaia Foundation (FBB, located at the Biofisika Institute).

The AINU (AI of the NUcleus) tool scans high-resolution images of cells. The images are obtained using a special microscopy technique called STORM, which creates an image that captures much finer details than conventional microscopes can see. The high-definition snapshots reveal structures at nanometer resolution.

A nanometer (nm) is one billionth of a meter, and a human hair is about 100,000 nm wide. AI can detect rearrangements within cells as small as 20 nm, which is 5,000 times smaller than the width of a human hair. These alterations are too small and subtle for human observers to detect using traditional methods alone.

“The resolution of these images is powerful enough for our AI to recognize specific patterns and differences with remarkable accuracy, including changes in the way DNA is arranged inside cells, allowing alterations to be spotted very quickly after they appear. We believe that one day, this type of information could save doctors valuable time to monitor diseases, personalize treatments and improve patient outcomes,” says Pia Cosma, ICREA research professor, co-corresponding author of the study and researcher at the Center for Genomic Regulation in Barcelona.

“Facial recognition” at the molecular level

AINU is a convolutional neural network, a type of AI specifically designed to analyze visual data such as images. Examples of convolutional neural networks include AI tools that allow users to unlock their smartphones with their faces, or others used by self-driving cars to understand and navigate environments by recognizing objects on the road.

In medicine, convolutional neural networks are used to analyze medical images such as mammograms or CT scans and identify signs of cancer that might go unnoticed by the naked eye. They can also help doctors detect abnormalities in MRI scans or X-ray images, leading to faster and more accurate diagnoses.

AINU detects and analyzes tiny structures inside cells at the molecular level. The researchers trained the model by feeding it nanometer-resolution images of the nucleus of many different cell types in different states. The model learned to recognize specific patterns in cells by analyzing how nuclear components are distributed and arranged in three-dimensional space.

For example, cancer cells have distinct changes in their nuclear structure compared to normal cells, such as alterations in the organization of their DNA or the distribution of enzymes in the nucleus. After training, AINU could analyze new images of cell nuclei and classify them as cancerous or normal based solely on these features.

The nanoscale resolution of the images allowed the AI ​​to detect changes in a cell’s nucleus as early as one hour after infection with herpes simplex virus type 1. The model was able to detect the presence of the virus by spotting slight differences in DNA density, which occurs when a virus begins to alter the structure of the cell’s nucleus.

“Our method allows us to detect cells infected by a virus very quickly after the onset of the infection. Normally, doctors take time to detect an infection because they rely on visible symptoms or larger changes in the body. But with AINU, we can immediately see tiny changes in the cell nucleus,” explains Ignacio Arganda-Carreras, co-corresponding author of the study and associate researcher of Ikerbasque at the UPV/EHU and affiliated with the FBB-Biofisika Institute and the DIPC of San Sebastian/Donostia.

“Researchers can use this technology to see how viruses affect cells almost immediately after entering the body, which could help develop better treatments and vaccines. In hospitals and clinics, AINU could be used to quickly diagnose infections from a simple blood or tissue sample, making the process faster and more accurate,” adds Limei Zhong, co-first author of the study and a researcher at Guangdong Provincial People’s Hospital (GDPH) in Guangzhou, China.

Laying the Foundations for Clinical Preparation

Researchers must overcome significant limitations before the technology is ready for testing or deployment in clinical settings. For example, STORM images can only be taken with specialized equipment typically found only in biomedical research labs. Setting up and maintaining the imaging systems required by AI represents a significant investment in equipment and technical expertise.

Another limitation is that STORM imaging typically analyzes only a few cells at a time. For diagnostic purposes, especially in clinical settings where speed and efficiency are crucial, physicians would need to capture a much higher number of cells in a single image to be able to detect or monitor disease.

“There are many rapid advances in STORM imaging, which means that microscopes could soon be available in smaller or less specialized labs, and eventually, even in clinics. Accessibility and throughput limitations are easier problems to solve than we previously thought, and we hope to be able to perform preclinical experiments soon,” says Dr. Cosma.

Although clinical benefits will be years away, AINU is expected to accelerate scientific research in the near term. Researchers have found that the technology can identify stem cells with high accuracy. Stem cells can develop into any type of cell in the body, an ability known as pluripotency. Pluripotent cells are studied for their potential to help repair or replace damaged tissue.

AINU can make the process of detecting pluripotent cells faster and more accurate, helping to make stem cell therapies safer and more effective.

“Current methods for detecting high-quality stem cells rely on animal testing. However, all our AI model needs to work is a sample stained with specific markers that highlight key nuclear features. In addition to being simpler and faster, it can accelerate stem cell research while contributing to the reduction of the use of animals in science,” explains Davide Carnevali, first author of the study and researcher at CRG.