Anne van Dalen

224 I Chapter 8 Jermyn et al. found that ANNs are able to improve the detection of invasive brain cancer cells by overcoming the negative impact of spectral artifacts. Despite the inclusion of light artifacts, ANNs keep the detection of invasive cancer cells at almost the same level, improving sensitivity by 19% and specificity by 7% compared to the standard technique. 26 Intraoperative Video Analysis Video data of laparoscopic procedures are used for both education and quality improvement purposes. In order to decrease the required time for analysis and review of video data, Hashimoto et al. investigated the possibility of automatic video segmentation using CV and ML techniques. Their research demonstrated that CV and ML techniques were able to differentiate between specific steps of laparoscopic surgery procedures with an accuracy of 82% ± 4%. 25 Workflow Recognition The long-term vision of Padoy 17 is to develop a surgical control tower (SCT) that, using AI, can monitor and support many processes, providing overall awareness of what is happening in the OR. Key for such an SCT is the requirement of an AI system that can recognize the surgical workflow and is aware of the surgical context. Workflow is often described as the sequence of tasks, interactions, or other processes through which a piece of work passes from initiation to completion. 31 In their review, Padoy 17 researched several recent ML and deep learning applications that can add to the workflow recognition system. These applications include phase recognition, tool detection and localization, and human detection and pose estimation and are described below. 17 Phase Recognition Phase recognition, the task of instantly determining the current phase of surgery at any time t from video data, was researched both in laparoscopic videos and external videos. In laparoscopic videos, a study showed that the combination of a CNN and a recurrent neural network (RNN) was able to recognize the different phases automatically and in real time, with an accuracy of 86%. In a study using external videos, a combination of a