SW prostředí pro automatizovanou analýzu retinálních lézí pro systémy Clarity RetCam 3 a PHOENIX Icon

Abstract

Retinopathy of prematurity is currently one of the most common causes of visual impairment in children. One of the possible manifestations of the disease is retinal lesions or hemorrhages. The aim of this thesis is to test an algorithm for segmenting hemorrhages on clinical data from Clarity RetCam 3 and Icon PHOENIX retinal cameras and then evaluate the segmentation efficiency between them. The final step is to implement the algorithm in a software environment that analyzes retinal lesions. The algorithm uses a semi-automatic iterative active contour method for lesion segmentation. Evaluation of the effectiveness of the segmentation algorithm was performed based on a quantitative measure of difference to gold standards. Based on statistical analyses, significant differences in segmentation scores were found in accuracy, sensitivity and specificity between retinal cameras. The largest difference in efficiency was observed in algorithm sensitivity, where Clarity RetCam 3 achieved a value of 85.63% compared to 81.79% for Icon PHOENIX. Geometric similarity of lesions and intensity spectrum showed an advantage for Clarity RetCam 3, with the largest difference in efficiency found in the variance of values, where Clarity RetCam 3 achieved a value of 34.54% compared to 51.75% for Icon PHOENIX.