Artificial intelligence based semi-automatic segmentation for orbital tumor preoperative modeling

With the increasing utilization of endoscopic approaches for primary benign orbital tumor (PBOT) surgery, otolaryngologists and ophthalmologists are challenged with determining candidacy for endoscopic resection based on preoperative imaging. Our objective was to develop a semi-automatic modality for three-dimensional reconstruction of primary benign orbital tumors to shed light on their resectability.

Patients with tumors from each ORBIT (Orbital Resection by Intranasal Technique) class were identified. Patients' preoperative imaging was obtained, and 3D slicer was utilized to segment the tumor with key anatomical structures, creating a three-dimensional model for each patient.

A model was generated for all five ORBIT classes. These models demonstrated the key findings differentiating classes such as whether a tumor is intraconal vs. extraconal, anterior or posterior to the intersection of the ophthalmic artery and optic nerve, or extends into the pterygopalatine and/or infratemporal fossa through inferior orbital fissure or through the superior orbital fissure.

Our study utilized 3DSlicer to semi-automatically develop three-dimensional reconstructions of each class of benign orbital tumors as designated by ORBIT criteria. These reconstructions may provide additional quantitative and qualitative data regarding these tumors, their resectability, and the potential surgical approaches required.