The AENEAS Project: Intraoperative Anatomical Guidance Through Real-Time Landmark Detection Using Machine Vision.

Area of Science:

• Neurosurgery
• Computer Vision
• Artificial Intelligence

Background:

• Microsurgical procedures, such as the pterional trans-Sylvian approach, require precise identification of anatomical structures.
• Intraoperative orientation can be challenging, potentially leading to increased surgical variability and risk.

Purpose of the Study:

• To evaluate the performance of a deep learning object detection model (YOLOv7x) for identifying key anatomical landmarks in the complex pterional trans-Sylvian approach.
• To assess the feasibility of machine vision techniques for anatomical recognition in neurosurgery.

Main Methods:

• A deep learning object detection model (YOLOv7x) was developed and trained on 5307 labeled frames from 78 surgical videos.
• Anatomical targets including optic nerves, internal carotid artery, dura, and cortical gyri were annotated.
• Model performance was evaluated using 5-fold cross-validation with average precision (AP50) as the primary metric.

Main Results:

• The model demonstrated promising detection accuracy for deep structures, achieving an AP50 of 0.73 for the optic nerve and 0.67 for the internal carotid artery.
• Detection of superficial structures like dura and cortical gyri showed lower precision (AP50 range: 0.25-0.45).
• Performance variability was attributed to anatomical complexity and data limitations.

Conclusions:

• Machine vision-based anatomical detection is feasible in complex neurosurgical settings.
• High accuracy for deep anatomical landmarks validates the approach, despite challenges with less distinct structures.
• This technology has the potential to enhance intraoperative orientation and standardize surgical practice.