2025 Proffered Presentations
S332: RESECTION OF PITUITARY ADENOMAS WITH SUBARACHNOID INVASION: ARTIFICIAL INTELLIGENCE ANALYSIS OF SURGICAL TECHNIQUE
Jonathan B Lamano, MD, PhD; Ana Sofia Alvarez, MD; Dhiraj J Pangal, MD; Michael T Chang, MD; Juan Carlos Fernandez-Miranda, MD; Stanford University
Introduction: A subset of pituitary adenomas invade through the suprasellar arachnoid plane and encase vital neurovascular structures. Unlike the resection of adenomas that respect the arachnoid plane, the resection of lesions with subarachnoid invasion requires a modified surgical technique to avoid inadvertent injury to involved neurovascular structures. This technique involves an expanded bony exposure, internal debulking, and meticulous, microsurgical, extracapsular dissection to safely detach the tumor from associated vascular structures and nerves. Surgical video analysis with artificial intelligence algorithms provides a mechanism to quantify and delineate these nuances in surgical technique.
Methods: Surgical videos of patients undergoing an endoscopic endonasal approach for pituitary adenoma resection from 2022-2023 were reviewed. Subarachnoid invasion was identified in 2 patients. For comparison, 6 patients with comparable tumor sizes were also identified. Surgical videos were uploaded to the Surgical Data Science Collective and analyzed using built-in machine learning and computer vision modules for instrument detection. Analysis was performed for the duration of tumor resection, defined from dural opening to completion of tumor resection. Instrument use statistics, timelines, and heatmaps were compared across patients with and without subarachnoid invasion.
Results: Tumor sizes across patients with and without subarachnoid invasion were matched, without significant difference in tumor diameter (26.2 mm versus 22.3 mm, p > 0.05). Length of surgical resection was not statistically different across tumors with and without subarachnoid invasion (96 minutes versus 80 minutes, p > 0.05). Instrument detection analysis identified tear-drop suction, microdissectors, microscissors, and pituitary grasper with a range of 70-99% confidence in detection. Resection of tumors with subarachnoid invasion was associated with an increased duration of use of the microdissectors (18.1% versus 10.3% of total surgical resection time, p = 0.01) and longer average duration of use per instrument appearance (11 seconds versus 9 seconds per appearance, p = 0.01). A trend towards increased percentage of microdissector appearances was also identified during resection of tumors with subarachnoid invasion (22.8% versus 14.4% of total instrument appearances, p = 0.06). No significant differences in instrument percentage use, average use duration, or percentage appearance were identified with the suction, microsscissors, and grasper across tumors with and without subarachnoid invasion (p > 0.05). Analysis of instrument timelines identified more consistent and prolonged use of the microdissectors throughout the duration of tumor resection when subarachnoid invasion was present. In addition, instrument heatmaps demonstrated increased central and circumferential suction and microdissector use during resection of adenomas with subarachnoid invasion.
Conclusion: Surgical resection of pituitary adenomas with subarachnoid invasion is associated with the risk of severe complications, primarily involving vascular injury and subsequent infarction. A modified surgical technique involving meticulous extracapsular dissection is necessary to avoid injury to involved neurovascular structures. Machine learning and computer vision analyses identified increased utilization of the microdissector, in addition to increased central and circumferential suction and microdissector use during resection of adenomas with subarachnoid invasion. This serves as an initial step in quantitatively and qualitatively delineating the nuances in the approach to resection of adenomas with subarachnoid invasion.
