Objective: To provide an overview of ML models and data streams utilized for automated surgical phase recognition. Background: Phase recognition identifies different steps and phases of an operation. ML is an evolving technology that allows analysis and interpretation of huge data sets. Automation of phase recognition based on data inputs is essential for optimization of workflow, surgical training, intraoperative assistance, patient safety, and efficiency. Methods: A systematic review was performed according to the Cochrane recommendations and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement. PubMed, Web of Science, IEEExplore, GoogleScholar, and CiteSeerX were searched. Literature describing phase recognition based on ML models and the capture of intraoperative signals during general surgery procedures was included. Results: A total of 2254 titles/abstracts were screened, and 35 full-texts were included. Most commonly used ML models were Hidden Markov Models and Artificial Neural Networks with a trend towards higher complexity over time. Most frequently used data types were feature learning from surgical videos and manual annotation of instrument use. Laparoscopic cholecystectomy was used most commonly, often achieving accuracy rates over 90%, though there was no consistent standardization of defined phases. Conclusions: ML for surgical phase recognition can be performed with high accuracy, depending on the model, data type, and complexity of surgery. Different intraoperative data inputs such as video and instrument type can successfully be used. Most ML models still require significant amounts of manual expert annotations for training. The ML models may drive surgical workflow towards standardization, efficiency, and objectiveness to improve patient outcome in the future. Registration PROSPERO: CRD42018108907

Image-based tracking of medical instruments is an integral part of surgical data science applications. Previous research has addressed the tasks of detecting, segmenting and tracking medical instruments based on laparoscopic video data. However, the proposed methods still tend to fail when applied to challenging images and do not generalize well to data they have not been trained on. This paper introduces the Heidelberg Colorectal (HeiCo) data set - the first publicly available data set enabling comprehensive benchmarking of medical instrument detection and segmentation algorithms with a specific emphasis on method robustness and generalization capabilities. Our data set comprises 30 laparoscopic videos and corresponding sensor data from medical devices in the operating room for three different types of laparoscopic surgery. Annotations include surgical phase labels for all video frames as well as information on instrument presence and corresponding instance-wise segmentation masks for surgical instruments (if any) in more than 10,000 individual frames. The data has successfully been used to organize international competitions within the Endoscopic Vision Challenges 2017 and 2019.

Objective: To depict and analyze learning curves for open, laparoscopic, and robotic pancreatoduodenectomy (PD) and distal pancreatectomy (DP). Background: Formal training is recommended for safe introduction of pancreatic surgery but definitions of learning curves vary and have not been standardized. Methods: A systematic search on PubMed, Web of Science, and CENTRAL databases identified studies on learning curves in pancreatic surgery. Primary outcome was the number needed to reach the learning curve as defined by the included studies. Secondary outcomes included endpoints defining learning curves, methods of analysis (statistical/arbitrary), and classification of learning phases. Results: Out of 1115 articles, 66 studies with 14,206 patients were included. Thirty-five studies (53%) based the learning curve analysis on statistical calculations. Most often used parameters to define learning curves were operative time (n = 51), blood loss (n = 17), and complications (n = 10). The number of procedures to surpass a first phase of learning curve was 30 (20–50) for open PD, 39 (11–60) for laparoscopic PD, 25 (8–100) for robotic PD (P = 0.521), 16 (3–17) for laparoscopic DP, and 15 (5–37) for robotic DP (P = 0.914). In a three-phase model, intraoperative parameters improved earlier (first to second phase: operating time –15%, blood loss –29%) whereas postoperative parameters improved later (second to third phase: complications –46%, postoperative pancreatic fistula –48%). Studies with higher sample sizes showed higher numbers of procedures needed to overcome the learning curve (rho = 0.64, P < 0.001). Conclusions: This study summarizes learning curves for open-, laparoscopic-, and robotic pancreatic surgery with different definitions, analysis methods, and confounding factors. A standardized reporting of learning curves and definition of phases (competency, proficiency, mastery) is desirable and proposed.

Background: The role of minimally invasive distal pancreatectomy is still unclear, and whether robotic distal pancreatectomy (RDP) offers benefits over laparoscopic distal pancreatectomy (LDP) is unknown because large multicentre studies are lacking. This study compared perioperative outcomes between RDP and LDP. Methods: A multicentre international propensity score-matched study included patients who underwent RDP or LDP for any indication in 21 European centres from six countries that performed at least 15 distal pancreatectomies annually (January 2011 to June 2019). Propensity score matching was based on preoperative characteristics in a 1 : 1 ratio. The primary outcome was the major morbidity rate (Clavien–Dindo grade IIIa or above). Results: A total of 1551 patients (407 RDP and 1144 LDP) were included in the study. Some 402 patients who had RDP were matched with 402 who underwent LDP. After matching, there was no difference between RDP and LDP groups in rates of major morbidity (14.2 versus 16.5 per cent respectively; P = 0.378), postoperative pancreatic fistula grade B/C (24.6 versus 26.5 per cent; P = 0.543) or 90-day mortality (0.5 versus 1.3 per cent; P = 0.268). RDP was associated with a longer duration of surgery than LDP (median 285 (i.q.r. 225–350) versus 240 (195–300) min respectively; P < 0.001), lower conversion rate (6.7 versus 15.2 per cent; P < 0.001), higher spleen preservation rate (81.4 versus 62.9 per cent; P = 0.001), longer hospital stay (median 8.5 (i.q.r. 7–12) versus 7 (6–10) days; P < 0.001) and lower readmission rate (11.0 versus 18.2 per cent; P = 0.004). Conclusion: The major morbidity rate was comparable between RDP and LDP. RDP was associated with improved rates of conversion, spleen preservation and readmission, to the detriment of longer duration of surgery and hospital stay.

The value of virtual reality (VR) simulators for robot-assisted surgery (RAS) for skill assessment and training of surgeons has not been established. This systematic review and meta-analysis aimed to identify evidence on transferability of surgical skills acquired on robotic VR simulators to the operating room and the predictive value of robotic VR simulator performance for intraoperative performance. MEDLINE, Cochrane Central Register of Controlled Trials, and Web of Science were searched systematically. Risk of bias was assessed using the Medical Education Research Study Quality Instrument and the Newcastle–Ottawa Scale for Education. Correlation coefficients were chosen as effect measure and pooled using the inverse-variance weighting approach. A random-effects model was applied to estimate the summary effect. A total of 14 131 potential articles were identified; there were eight studies eligible for qualitative and three for quantitative analysis. Three of four studies demonstrated transfer of surgical skills from robotic VR simulators to the operating room measured by time and technical surgical performance. Two of three studies found significant positive correlations between robotic VR simulator performance and intraoperative technical surgical performance; quantitative analysis revealed a positive combined correlation (r = 0.67, 95 per cent c.i. 0.22 to 0.88). Technical surgical skills acquired through robotic VR simulator training can be transferred to the operating room, and operating room performance seems to be predictable by robotic VR simulator performance. VR training can therefore be justified before operating on patients. This systematic review and meta-analysis presents current evidence on transferability of surgical skills acquired on robotic VR simulators to the real operating room, and on the predictability of intraoperative performance by robotic VR simulator performances. The limited data currently available support the use of robotic VR simulators for surgical skill acquisition and assessment. This systematic review and meta-analysis presents current evidence on transferability of surgical skills acquired on robotic VR simulators to the real operating room, and on the predictability of intraoperative performance by robotic VR simulator performances. The limited data currently available support the use of robotic VR simulators for surgical skill acquisition and assessment.