Romy C. van Jaarsveld

Deep learning-based surgical vision systems achieve strong performance in semantic segmentation and phase recognition, but their black-box nature limits traceability in safety-critical clinical settings. Prototype-based networks offer an interpretable alternative by grounding predictions in learned visual exemplars, yet their suitability for surgical video understanding remains insufficiently characterized. We adapted a prototype-based architecture to two surgical datasets, laparoscopic cholecystectomy and robot-assisted minimally invasive esophagectomy (RAMIE), and benchmarked it against conventional baselines. We evaluated a segmentation-only setting, in which prototype size and capacity were ablated, and a multitask setting, in which three strategies for coupling prototype learning to semantic segmentation and surgical phase recognition were compared. Prototype-based models underperformed the conventional baselines across both tasks and datasets. In the segmentation-only setting, the selected prototype configurations achieved Dice scores of 72.23% on Cholecystectomy and 72.07% on RAMIE, compared with 74.35% and 74.02% for the corresponding conventional baselines, and showed weaker boundary agreement. In the multitask setting, the best prototype strategy recovered competitive segmentation performance but remained 6–12 F1 points below the conventional baseline for phase recognition. Qualitatively, prototype activation maps exposed intra-structure decompositions and contextual cues that are not directly available from black-box baselines. Prototype networks provide spatially traceable evidence for surgical scene understanding, but currently trade interpretability for reduced boundary precision and phase-recognition performance. These findings motivate future work on scene-level and temporally aware prototypes for explainable surgical AI.

Background: Curative treatment of resectable esophageal cancer comprises neoadjuvant chemoradiotherapy and esophagectomy. Robot-assisted minimally invasive esophagectomy (RAMIE) is the preferred technique; however, learning RAMIE is challenging due to the complex chest anatomy, patient positioning and zoomed-in camera view. Computer-aided anatomy recognition holds promise for improving surgical navigation. This study aims to develop real-time anatomy recognition and surgical phase recognition algorithms for the thoracic part of RAMIE using deep learning and to understand the challenges of current state-of-the-art algorithms. Methods: A retrospective single-center cohort study was conducted on prospectively collected RAMIE videos at University Medical Center Utrecht, The Netherlands. Two datasets were created: an anatomy segmentation dataset with 1504 frames from 53 videos, annotated for eight anatomical classes and four surgical instruments; and a surgical phase dataset, with 38 videos labeled with thirteen distinct phases. Several deep learning models were trained and tested using both datasets. Results: The SegNeXt model achieved the most accurate segmentations with an overall overlap score of 0.72 for all classes, of which the surgical instruments were best detectable. The best surgical phase recognition model achieved an overall accuracy of 82.8% and revealed loss of accuracy during phase transitions. Class imbalance affected both datasets which led to less frequent appearing classes performing lowest. Conclusions: This study identified suitable models for computer-aided, real-time anatomy and surgical tool segmentation and surgical phase recognition for the thoracic part of RAMIE. Acceptable performance was achieved while both datasets are highly complex. Performance is expected to improve by further expanding and diversifying the datasets.

Background: Transthoracic esophagectomy (TTE) is the preferred approach for curative esophageal cancer. However, concomitant single-lung ventilation limits its use in patients with compromised cardio-pulmonary function. Transhiatal esophagectomy (THE) grants two-lung ventilation yet offers a limited mediastinal lymphadenectomy and is only suitable for distal tumors. A promising alternative is robot-assisted cervical esophagectomy (RACE). Initial studies show its feasibility, yet, the mediastinal lymph node yield (LNY) is unknown. The aim of this study is to describe the accessibility of the lymph node stations (LNS) and to analyze to which extent the mediastinal LNS can be resected during the cervical phase of RACE. Methods: Experienced robotic upper-gastrointestinal (GI) surgeons performed cervical esophagectomy and mediastinal lymphadenectomy on 4 cadavers using a “da Vinci Xi” surgical system through a left cervical approach. Target LNS were 4R, 4L, 5 and 7, as defined by Naruke. Resected tissues were categorized as primary—esophagus and lymph nodes (LNs) resected through RACE- or secondary—residual LNs post-RACE- resected tissue and were analyzed by experienced pathologists. LNs were counted for every station. Results: Station 4R yielded a median of 16 LNs primarily and 12 secondarily; station 4L: 17 primarily, 0 secondarily; station 5: 4 primarily, 7 secondarily; and for station 7: 9 primarily and 5 secondarily. Conclusions: This study shows that the RACE procedure delivers a limited mediastinal lymphadenectomy. LNS 4L can be fully resected. Stations 4R, 5 and 7 could be partially resected. The RACE procedure may, however, be a viable alternative for patients who cannot undergo a TTE.

Background: During thoracoscopic esophageal resection, while performing the supracarinal lymphadenectomy along the left recurrent laryngeal nerve (LRLN) from the aortic arch to the thoracic apex, we observed a not previously described bilayered fascia-like structure, serving as prolongation of the already known mesoesophagus. Methods: We retrospectively evaluated 70 consecutively unedited videos of thoracoscopic interventions on esophageal resections for cancer, in order to determine the validity of this finding and to describe its utility for performing a systematic and more accurate dissection of the LRLN and its adequate lymphadenectomy. Results: After mobilization of the upper esophagus from the trachea and tilting the esophagus by means of two ribbons, a bilayered fascia was observed between the esophagus and the left subclavian artery in 63 of the 70 patients included in this study. By opening the right layer, the left recurrent nerve became visualized and could be dissected free in its whole trajectory. Vessels and branches of the LRLN were divided between miniclips. Mobilizing the esophagus to the right, the base of this fascia could be found at the left subclavian artery. After dissecting and clipping the thoracic duct, complete lymphadenectomy of 2 and 4L stations could be performed. Mobilizing the esophagus in distal direction, the fascia continued at the level of the aortic arch, where it had to be divided in order to mobilize the esophagus from the left bronchus. Here, a lymphadenectomy of the aorta-pulmonary window lymph nodes (station 8) can be performed. It seems that from there the fascia continued without interruption with the previously described mesoesophagus between the thoracic aorta and the esophagus. Conclusions: Here we described the concept of the supracarinal mesoesophagus on the left side. Applying the description of the mesoesophagus will create a better understanding of the supracarinal anatomy, leading to a more adequate and reproducible surgery.