Anatomical Landmarks Detection for Laparoscopic Surgery Based on Deep Learning Technology

The modern trend in medical video systems development is using for analysis and visualization images of different modalities such as narrow band images or fluorescents images. The method of landmarks detection for laparoscopic images based on deep learning technologies with application additional information obtained from fluorescent images is proposed. The main feature is small data base of images obtained in white light for CNN training and extracting additional information from high-intensity fluorescence in ICG laparoscopy by methods of traditional machine learning. The combination of CNN approach and machine learning approach for fluorescent information using allows to increase the quality of landmarks segmentation in comparing with methods based only on CNN. Proposed method was tested on real laparoscopic images.