AI model detects brain tumors with high precision using epigenetic fingerprints

The MRI image shows a brain tumor in an uncomfortable position, – and a brain biopsy will result in high risks to the patient, who had consulted us due to double vision. Situations such as this case discussion, referred to as an example, in a multi -scientific team of cancer medicine experts, has sparked researchers at Charité – Universitätsmedizin Berlin, along with collaborators to seek new diagnostic procedures. The result is an AI model. The model uses specific features in the genetic material of the tumors – the epigenetic fingerprint, which is taken for example from the cerebrospinal fluid, among others. As the group shows in the magazine Cancer natureThe new model classifies tumors quickly and very reliable.

Today, many more types of tumors are known than the organs from which they arise. Each volume has its own characteristics, certain tissues, growth rates and metabolic particularities. However, types of tumors with similar molecular characteristics can be grouped. The treatment of individual disease depends decisively on the type of tumor. New, targeted treatments treat certain structures of cancer cells or prevent their signaling pathways to stop the development of pathological tissue. Chemotherapy can be selected according to the type of tumor and their dosage to be adjusted accordingly. Particularly in the case of rare types of tumors, it may be possible to pursue innovative therapies as part of the studies.

In the context of the increasingly personalized, rapid development of cancer medicine, the exact diagnosis in a certified tumor center is the way forward for successful treatment. “

Professor Martin E. Kreis, Head Physician in Charité

While a comprehensive molecular, cellular and functional analysis of a tissue -based tumor provides the necessary information, doctors also face cases where it is not possible or too dangerous to extract tissue samples from the tumor. In addition, even a histological examination in itself is not able to provide an accurate diagnosis such as the new AI model.

Looking at the genome instead of the web

A method has been created for the characterization of brain tumors not based on conventional tiny diagnostics, but on modifications to the genetic material of the tumor, epigenetic characteristics. It is part of the memory of each cell and determine which parts of genetic information are read and when. “Hundreds of thousands of epigenetic modifications act as switches and switches disabled for individual gene sections. Their standards are a unique, undisputed fingerprint,” explains Dr. Philipp Euskirchen, a scientist at the German Consortium, publication. “In volume cells, epigenetic information changes in a characteristic way. Based on their profiles, we can differentiate tumors and classify them.” In the case of brain tumors, even a sample of cerebrospinal fluid is adequate in some cases and can be relatively easily obtained – completely surgery.

In order to compare an unknown fingerprint with thousands of well -known fingerprints of different cancers and to assign them to a specific type of volume, mechanical learning methods, that is, artificial intelligence, as the data is very extensive and complex. In addition, different methods of DNA have been applied in the past. In addition, epigenetic analyzes are usually limited to defined standards and gene segments that are typical of individual types of tumors. “As a result, our goal was to develop a model that accurately classifies tumors, even if they are only based on parts of the entire epigemia or profiles at the Berlin Institute with different techniques and varied degrees of accuracy,” such as biopathic.

Reliable and detectable

A recently developed AI model goes by the name Crossnn, whose architecture is based on a simple neuronal network. The model was trained with a large number of reference volumes and then tested in more than 5,000 volumes. “Our model allows us to have a very accurate diagnosis of brain tumors in 99.1 % of all cases and is more accurate than AI solutions at work to date,” as related to Philipp Euskirchen. “In addition, we have been able to train an AI model in the same way that it can differentiate more than 170 types of tumors from all organs, while achieving the accuracy of 97.8 %. The decisive factor for future approvals in the clinical application is that models are fully explained.

The molecular fingerprint that the AI ​​receives for determination may come from a tissue sample or from body fluids. In the case of specific brain tumors, the Department of Neuropathology in Charité already offers non -invasive diagnostics based on cerebrospinal fluid, known as liquid biopsy. This allows the diagnosis to be made without a stressful function, also in difficult situations. The patient who consulted us with double vision was one of the beneficiaries. “We examined the cerebrospinal fluid using the sequence of nanopers, a new, very fast and effective form of genetic analysis.

Crossnn in clinical trials

The accuracy of the methodology even received the researchers by surprise. “Although the architecture of the AI ​​model is much simpler than previous approaches and therefore remains explained, it offers more accurate predictions and therefore more diagnostic certainty,” says Sören Lukassen. Together with the German Cancer Consortium (DKTK), the research team is therefore planning clinical trials with Crossnn and eight DKTK positions in Germany. In addition, intra -operative use should also be tested. The goal is to transfer the exact and comparatively cheap tumor determination based on DNA samples to care routine.