AI system for brain tumour diagnosis developed at Lobachevsky University
Artificial intelligence (AI) models based on machine learning can provide highly accurate diagnosis of glioma, one of the most aggressive types of brain tumour. The system determines the subtype of the tumour, predicts the patient's survival, and will eventually aid in choosing the optimal treatment tactics. The diagnosis relies on the activity of 13 key genes that are associated with three major subtypes of glioma: astrocytoma, oligodendroglioma, and glioblastoma.
"The classical approach to diagnosing gliomas does not always provide a quick and accurate determination of the type of tumour. Histological analysis and the search for specific mutations can lead to conflicting diagnoses, and even advanced MRI techniques may not provide sufficient information due to the variability of gliomas. Our tool uses transcriptome analysis to evaluate tumours, identifying which genes are "on" or "off" and allowing us to assess the activity levels of these genes in individual patients. This approach has the advantage of identifying the specific subtype of glioma, enabling doctors to choose the most appropriate treatment method," says Mikhail Ivanchenko, Director of the Biogerontology Research Institute at Lobachevsky University, who led the study.
Many AI models operate on a "black box" principle, which makes it hard for the doctors to understand the logic behind these models’ decision-making process. Nizhny Novgorod scientists have taken a different approach, using Explicable Artificial Intelligence (XAI) to address this issue. Thanks to additional algorithms, the system now shows exactly how the level of expression of each of the 13 significant genes affects the model's predictions.
"The answer will be different for each patient, even if the type of tumour is the same. The doctors can double-check the neural network's decision and either agree or disagree with it. They can also conduct additional research if needed. To ensure trust in AI and make informed clinical decisions, experts need to validate the diagnosis and prognosis. The opportunity to look inside the biological processes behind the prognosis opens up new perspectives for personalised medicine," comments Dr. Mikhail Ivanchenko.
According to scientists, this approach can currently be used primarily for laboratory scientific research. In the future, they plan to create clinical test systems based on this development to assess the level of gene expression, which would allow for accurate diagnoses to be made in a shorter time.
This research was carried out by scientists from the UNN Research Centre for Artificial Intelligence and the UNN Biogerontology Research Institute, and their findings were published in a special issue of the Cancers Journal.
