'The system is designed to support doctors in the most tedious activities and give them more time to analyse diagnostically significant features', says Przemysław Biecek, project manager and leader of the MI2.AI research team. The team deals with data processing and machine learning.

The expert explains in the release sent to PAP that when looking for similar images, the system (developed in the Xlungs project) can quickly review thousands of reference computed tomography scans, and in each scan it quickly analyses hundreds of images to precisely mark pathological changes and important anatomical features.

'Anatomical features measured by the system can be integrated with other diagnostic processes. Just like a blood test is the basis for diagnosing numerous diseases, such precise, fast and cheap dimensioning of changes in the chest can be a breakthrough in screening', believes the MI2.AI team leader. He adds that this tool can be integrated with systems already used in treatment, because it works with commonly accepted standards of medical documentation.

The system uses a huge number of lung CT images (as many as 40 thousand), developed by the MI2.AI research team from the Warsaw University of Technology in collaboration with the Polish Lung Cancer Group. The source of these images were CDs with CT scans of Polish patients from 2010-2018, including scans lung cancer screening tests. As a result, an artificial intelligence model was created based on the largest database of this type in the world (40 terabytes of data). It will support doctors in faster and more effective diagnosis of diseases. However, there may be many more such resources in Poland.

Several hundred million laboratory tests are performed in Poland every year, of which over 60 million are imaging tests, such as computed tomography (CT). According to the Collective Minds Radiology report, a single CT scan generates from 200 MB to 1 GB of data. An average-sized hospital generates several dozen terabytes (1 TB = 1024 GB) to several petabytes (1 PB = 1024 TB) of data per year in the form of imaging scans, laboratory results and medical records.

For over a decade, Poland has been building electronic medical records - an integrated system collecting patient health data. Since July 1, 2021, every doctor or office has been required to report medical events in it. Many medical facilities had already been collecting such data on their own even before that date.

'We often deal with a situation when a patient's treatment has already ended, and the results of their tests are still in the hospital or clinic database, metaphorically +gathering dust+ on the shelves', explains Marcin Luckner, head of the work carried out in the Xlungs project. 'However, even if the collected data are no longer applicable to a given disease case, comparing them with the results of other people struggling with the same disease can allow doctors to see certain patterns and regularities in the development of the disease, and improve its treatment in the future. Such an analysis of hundreds or thousands of test results is a very tedious and time-consuming work, but artificial intelligence algorithms can support us in this area', he argues.

According to the scientists from the Warsaw University of Technology, several thousand computer science students graduate in Poland every year, and half a billion e-prescriptions are issued every year. They believe that we therefore have a good chance of becoming a leader in the creation of AI-supported medical technologies. 'Polish data allows to better support local diagnostics than data obtained from China, for example. At the same time, their volume gives the potential to create world-class solutions', the experts from the Warsaw University of Technology emphasise.

The CTSegMate artificial intelligence model developed in the Xlungs project extracts key information from historical CT images, automates the process of describing the results and shortens the time of their analysis. It took the MI2.AI team three years to create it, and the processors involved in this task needed over 180,000 hours of calculations. The implementation of the project was possible thanks to funding from the Polish National Centre for Research and Development (the INFOSTRATEG I competition).

The members of the MI2.AI team are academics and students of two faculties of mathematics and computer science in Poland: at the University of Warsaw and the Warsaw University of Technology. They conduct scientific research in the field of artificial intelligence, work on the practical application of research results, and promote knowledge in their area of expertise. (PAP)

zbw/ zan/