Çevik K. K. (Executive)
TUBITAK Project, 2024 - 2025
Infrared thermography is an imaging modality used for clinical applications, such as evaluation of skin perfusion in burn patients, dermatological conditions, and also in plastic surgery. Tissue perfusion is an important post-operative issue, especially for plastic surgeons. Failure of operative microvascular connections following reconstructive free tissue transplantation can cause irreparable damage or death of the transplanted tissues if not detected early enough. This leads to the need for corrective surgery and may result in additional costs. Standard monitoring of features such as bleeding, color and temperature change and turgor (swelling, hardness) is performed continuously for 48-72 hours after the surgical procedure. However, factors such as tiredness of the personnel performing this assignment or working under an excessive workload may cause delays in the detection of negative outcomes. In this study, it is aimed to design and develop an automatic post-surgical device for continuous unattended monitoring of flap regions and early detection of possible complications. The proposed system will combine images from complementary detection methods such as Thermal IR (Infrared), RGB and laser imaging, detection and ranging (LIDAR), perform scoring, classification and segmentation with the help of artificial intelligence and deep learning algorithms, and ultimately provide a detection decision regarding tissue perfusion. In solving the flap compromise problem; Long-term memory (LSTM) will be used for flap score estimation, Convolutional Neural Network (CNN) will be used for flap compromise classification, and current semantic segmentation algorithms will be used for segmentation. For the preliminary tests of the system, Mask RCNN method, which is frequently used in the literature and has the feature of performing scoring, classification and semantic segmentation simultaneously in a single model, will also be used. The process of providing data to the project from the designed multi-modal imaging mechanism will be done with volunteers at the first stage. Arterial insufficiency will be created by applying a tourniquet to one arm of the volunteer for a total of 5 minutes. Using RGB, Thermal IR and LIDAR cameras, images of the relevant region will be taken simultaneously for each patient and added to the data set. Deep neural network architectures to be designed with the obtained data set will be tested and their success rates will be evaluated. For measuring the performance of the proposed system, performance metrics such as confusion matrix, accuracy, precision, F1 Score, mAP and IoU will be used. After the system is found successful with the created data sets, real-time tests will be performed on patients. It is thought that with the success of the system, human-induced errors in monitoring and detection of tissue perfusion will decrease and the maintenance cost will be reduced significantly.