An analysis was made of the use of neural network algorithms for the detection of colon polyps obtained during colonoscopy. The Kvasir-SEG image database was used to train and test deep machine learning algorithms. The networks YOLOv6, YOLOR, YOLOv7, YOLOv7X, YOLOv8 previously trained on the basis of MS COCO images were used as neural network architectures. Due to the small volume of images in the Kvasir-SEG database, data augmentation was used. As a result of applying the detection algorithms to the test set of endoscopic images, the highest values of the metrics AP@[0,25..0,75] equal to 98,4 and AP@0,50 equal to 98,6 were obtained for the neural network detector based on the YOLOv8 network. According to the results of comparing the proposed algorithms with analogues, the YOLOv8 assessment showed an increase in the results for the AP@[0.25..0.75] metric by 5.9 in searches with the previous model НАОО^YL. The results obtained can be used in the development of a video stream analysis system in an endoscopic system operating in real time during colonoscopy studies

Endoscopy, images, polyps, deep learning, YOLO neural network architecture

Khryashev V.V.,  ,  orcid.org/0000-0003-2978-2459,  P.G. Demidov Yaroslavl State University

Kotov N.V.,  ,  orcid.org/0000-0002-6207-473X,  P.G. Demidov Yaroslavl State University

Priorov A.L.,  ,  orcid.org/0000-0003-2984-340X,  P.G. Demidov Yaroslavl State University