Restricted connectivity neural network structure for organ recognition by analysis of endoscopic images during surgical operation

This paper designs a neural network (NN) based system for recognition of presence of an internal organ on colour images from endoscope during abdominal surgery. NN-based system proposed in the paper is capable of dividing them into two groups: with presence of the liver in the image on the screen and without it. Restricted connectivity structure of the network makes possible decomposition of the image during the analysis and significantly reduces the number of parameters thus making training easier, faster and more accurate. Moreover, it reduces calculation time when trained network is used and makes possible to use the proposed system for real time image analysis during the operation where reaction time is critically important. The effectiveness of the proposed NN-based system is demonstrated on simulations.