A team of scientists from Tomsk Polytechnic University (TPU) and Skoltech has developed a universal method based on artificial neural networks for accurately predicting key parameters of polymer scaffolds — biocompatible frameworks for tissue regeneration. This discovery significantly reduces the time and resources required for the development and optimization of such implants, which is critical for the accelerated implementation of personalized regenerative medicine.
Previously, finding the optimal conditions for scaffold production (such as fiber diameter and strength) required numerous expensive and lengthy experiments. Traditional statistical modeling methods, as the study showed, often proved to be inconsistent:
The research results showed that Box-Behnken design models failed to predict data on fiber diameter and tensile strength.
The new approach uses two-layer perceptrons that are capable of identifying complex dependencies even from small datasets of experimental data. This not only increases the accuracy of the forecast but also eliminates the need to conduct extensive laboratory tests each time.
Moreover, the developed method can be used not only for polycaprolactone scaffolds but also for various types of polymers from which scaffolds can be obtained by electrospinning.
Calculations have confirmed that neural network models significantly outperform previous methods in accuracy, especially in predicting mechanical strength — a key parameter for successful implantation. In the future, this opens the way to creating automated systems for designing scaffolds with specified properties, which will allow for faster and cheaper creation of effective solutions for restoring damaged tissues in patients.
The results of the work, supported by the Ministry of Science and Higher Education of the Russian Federation, have been published in the authoritative international journal Computers in Biology and Medicine.
Read more materials on the topic:
No diesel — it will run on oil: a prototype of biofuel based on rapeseed oil was created at NovSU