|Complications related to the misplacement and dislodging of endotracheal intu-bation can be lethal. A novel artificial intelligence system for endotracheal intu-bation confirmation is described, based on the processing of images acquired during the intubation and identification of carina images. The system is com-prised of a miniature metal oxide silicon sensor (CMOS) attached to the tip of a semi-rigid stylet connected to a digital signal processor (DSP) with an integrat-ed video acquisition component. Video images are acquired and processed by a deep learning algorithm implemented on the DSP. The algorithm calculates the probability that the image belongs to the carina model that was a-priori trained and saved, and used for the classification decision. System performance was as-sessed on intubation videos recorded from 10 subjects who underwent general anesthesia during elective operations. The videos were annotated by an anesthe-siologist, such that each video image was classified according to the anatomical position. A leave-one-case-out method was employed, such that in each itera-tion, images from one subject were used to train the system and estimate the ca-rina model, and the remaining images were used to assess system performance. This process was repeated 10 times such that each subject participated once in testing. The results showed that this fully automatic image recognition system confirmed the correct tube positioning in all cases, for a 100% success rate.|
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.