Penza State University Creates Unique Neural Network for Precise Heart Surgery Planning

Specialists from Penza State University (PSU) have developed the 3D-Corvasculograph program, which uses graph neural networks to model the three-dimensional structure of the heart's coronary arteries. This will allow surgeons to more accurately plan angioplasty — minimally invasive procedures to restore blood flow in the vessels.

The accuracy of predicting artery diameters reaches 97%, which reduces the risk of complications and opens the way to personalized cardiac surgery. The program will also be used in the educational process: students will be able to study anatomy on digital models, observing the effect of changes in vessel parameters on blood flow.

Angioplasty with stent placement helps save patients from heart attacks and strokes, but the success of the operation depends on an accurate understanding of the individual anatomy of the vessels. An error of even half a millimeter can lead to restenosis or rupture of the artery.

Ordinary neural networks cannot cope with such branched structures, so scientists have applied graph networks, which take into account the hierarchy of bifurcations and the relationships between them.

3D-Corvasculograph restores the original three-dimensional map of the coronary arteries before the development of pathology. To do this, it is enough to enter the initial diameter of the vessel, after which the neural network generates the entire network of arteries with thousands of branches, taking into account unique bifurcations — the branching points of the arteries.

Knowing the original anatomy, we can choose the optimal stent size and surgical strategy

Anna Sergienko, postgraduate student at the PSU Medical Institute

Data from 30 corrosion specimens of arteries from healthy people were used to train the neural network. A contrast agent was introduced into the vessels, CT scans were performed, and 957 bifurcations were manually measured.