Action Classification for Endoscopic Pituitary Adenoma Resection: A Consensus-Based Study

Introduction: Pituitary adenomas are common brain tumors that have a profound impact on patient morbidity, mortality, and quality of life. The transsphenoidal surgical approach to these adenomas demands high surgical skill, which significantly affects outcomes. However, the specific aspects of surgical performance that influence these outcomes are not well understood. The first step to tackling this problem is to understand how these surgeries are performed and compare the relationships between performance and outcomes. Operative workflow analysis enables this through the deconstruction of an operation into its fundamental building blocks, including its distinct phases, steps, tasks and actions. Phases (e.g., nasal approach) may be composed of steps (e.g., tumor resection) which are composed of tasks (e.g., removing tumor tissue) through specific actions. Actions are defined as the interaction of instruments (e.g., pituitary rongeur) with a target (e.g., tumor) through verbs (e.g., grasping). Expanding on our previous work analyzing the phases and steps of these operations, the current work strives to deliver a universal framework to analyze the tasks and actions involved in a pituitary adenoma resection. Such a framework may guide surgical training and facilitate the development of AI models that may offer intraoperative decision support or individualize postoperative care of pituitary patients.

Methods: We developed universal classification ontology for tasks and actions in the tumor resection step of endoscopic transsphenoidal surgery (eTSA) for pituitary adenomas, utilizing a multicentric, multidisciplinary, consensus-based approach. This ontology was informed by endoscopic videos from two major international pituitary centers. A panel of AI engineers and attending skull base surgeons was established. A detailed action classification system was created and refined through four rounds of review and consensus discussions, focusing on the operational definitions of instruments, targets, and verbs in surgical actions.

Results: The final framework, derived from the analysis of 17 primary endoscopic resections (8 microadenomas, 9 macroadenomas), incorporated 9 verbs, 16 instruments, and 7 targets, defining 6 critical tasks ([Table 1]). [Figs. 1], [2], and [3] depict common actions of the surgical step. Annotated video data between microadenoma and macroadenoma data resections was reviewed. The annotation framework was analyzed for its inter-annotator consistency, interrogating its reliability and clinical applicability.

Discussion and Conclusion: This work established a comprehensive, universally applicable task and action ontology for the tumor resection of pituitary adenomas via eTSA. This ontology facilitates the detailed annotation of surgical videos that may be used for surgical training and sets the stage for future AI applications in surgery that could offer real-time intraoperative assistance and enhance postoperative outcomes. The methodology applied in developing this ontology emphasizes the importance of a systematic, consensus-based approach to enhance the clinical applicability and universality of surgical annotations.

Table 1

Fig. 1 Creating plane between tumor and dura.

Fig. 2 Removal of tumor tissue.

Fig. 3 Mobilize tumor tissue.

Publication History

Article published online:
07 February 2025