Automated Biometric Quantification of Lower Eyelid Malposition Using YOLOv11

Abstract: This study introduces an automated computer vision framework designed to standardize the diagnosis of lower eyelid malposition (Retraction and Ectropion). Utilizing a custom-trained YOLOv11 deep learning architecture, the developed system performs real-time detection of periocular landmarks to calculate critical surgical metrics, specifically Margin Reflex Distance 2 (MRD2) and Palpebral Fissure Height (PFH). Problem Statement: Current clinical diagnostics rely on manual caliper measurements, which are prone to parallax error, reflex blepharospasm, and high inter-observer variability. This subjectivity often leads to "Undiagnosed Ptosis" or inadequate surgical planning for procedures such as Vertical Midface Lifting. Methodology: The proposed solution is a Python-based desktop application integrating OpenCV for image processing and Ultralytics YOLOv11 for inference. Key technical contributions include: Dynamic Resolution Scaling: An algorithm that adapts pixel-to-centimeter conversion based on image dimensions, eliminating the need for physical reference markers. Geometric Correction: A spherical projection correction factor (0.75) to account for the 3D curvature of the globe on 2D static images. Hardware Acceleration: A hybrid inference pipeline supporting CUDA-enabled GPUs for real-time analysis (