Vývoj algoritmu pro automatizovanou detekci ptózy na základě analýzy antropometrických parametrů extrahovaných z 3D skenů obličeje

Abstract

This master’s thesis focuses on the use of three-dimensional scanning and machine learning methods for the automated detection of ptosis. The aim of the thesis was to design, implement, and experimentally validate a classification model capable of distinguishing between a healthy eye and an eye affected by ptosis based on anthropometric parameters extracted from 3D facial scans. The theoretical part of the thesis addresses the anatomy and pathology of the eyelids, the principles of 3D scanning, and an overview of selected machine learning methods suitable for classification tasks. It also includes a literature review of current approaches to ptosis diagnosis, highlighting the limitations of 2D image analysis and emphasizing the benefits of using spatial information. The practical part focuses on the creation of a custom dataset based on 3D scans of the periocular region, from which relevant anthropometric parameters were extracted and used as input for classification models. These models were trained, optimized, and compared, and their performance was further evaluated against blinded physician assessment. The results demonstrate that the use of 3D data in combination with machine learning enables high classification accuracy and provides more objective approach to ptosis evaluation. The proposed approach thus represents a promising tool for supporting clinical diagnostics and for the further development of automated methods in the field of ophthalmology.