Development of an algorithm for predicting the results of laser surgical procedures based on ensemble methods of machine learning
Victoria V. Suchkova, 1,2 Dmitry I. Ryabkin, 1,2 Alexander Yu. Gerasimenko, 1,2
1 I. M. Sechenov First Moscow State Medical University, Moscow, Russia
2 National Research University of Electronic Technology MIET, Zelenograd, Moscow, Russia
Abstract
Laser treatment allows achieving wound waterproofing, sealing small vessels and thereby reducing the duration of bleeding, there is also no suture compression of tissues and their marginal necrosis, so the tissue heals without epithelial overgrowth and tissue scarring. Since laser methods are non-contact, the risk of infection in the wound is minimal. Compared to other non-contact methods (ultrasound and electric welding), the use of fiber light guides in the case of laser methods allows for endoscopic and laparoscopic surgeries. Despite significant advances in the experimental study of laser treatment methods, the lack of a high degree of repeatability of results hinders the standardization of clinical application of lasers in regenerative medicine. The use of machine learning techniques to predict laser soldering outcomes will help ensure the safety of tissue repair surgery.
The research has developed a program for estimating the probability of success of the procedure of laser repair of biological tissues based on ensemble methods of machine learning (random forest, gradient bousting and its variations: lightGBM, XGboost. The program performs the function of data preprocessing, allows visualizing the initial data set with the description of laser repair operations, estimating the contribution of each parameter to the final strength value, predicting the strength value of the formed welds, as well as evaluating the accuracy of the obtained values. The following parameters are used as input parameters for strength prediction in this program: characteristics of the applied laser radiation (wavelength, power, duration, temperature, etc.), component composition of auxiliary substances (adhesives, membranes, solders), type of biological tissue.
To experimentally validate the strength algorithm, a series of ex vivo and in vivo experiments were conducted on laser reconstruction of biological tissues (skin) using the most commonly used bioorganic solders (bovine serum albumin, indocyanine green) and a laser complex with the ability to maintain a set temperature in the weld formation zone.
File with abstract
Speaker
Victoria V. Suchkova
1 I. M. Sechenov First Moscow State Medical University, Moscow, Russia
Russia
Discussion
Ask question
