Premium
Multimarkers for diabetic retinopathy screening
Author(s) -
CSUTAK A,
TOROK ZS,
TUKACS E,
MAROSSZABO ZS,
CSOSZ E,
BERTA A,
MOLNAR AM,
TOZSER J,
NAGY V,
DOMOKOS B,
HAJDU A
Publication year - 2012
Publication title -
acta ophthalmologica
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.534
H-Index - 87
eISSN - 1755-3768
pISSN - 1755-375X
DOI - 10.1111/j.1755-3768.2012.s037.x
Subject(s) - diabetic retinopathy , machine learning , computer science , artificial intelligence , fundus (uterus) , parameterized complexity , biomarker , usability , algorithm , medicine , diabetes mellitus , ophthalmology , human–computer interaction , biochemistry , chemistry , endocrinology
Purpose The aim of the project was to develop a methodology for diabetic retinopathy (DR) screening based on the examination of tear fluid biomarker changes. To evaluate the usability of protein biomarkers for pre‐screening purposes different approaches and machine learning algorithms were used. Methods All persons involved in the study had diabetes. DR was diagnosed by capturing 7‐field fundus images. 165 eyes were examined, 55 were diagnosed healthy and 110 images showed signs of DR. Tears were taken from all eyes and state‐of‐the‐art nano‐HPLC coupled ESI‐MS/MS mass spectrometry protein identification was performed on them. Applicability of protein biomarkers was evaluated by six different optimal parameterized machine learning algorithms. Results Out of the six identified machine learning algorithms, result of the Recursive Partitioning proved to be the most accurate. The performance indicators of the system applying the above algorithm indicated 74 % sensitivity and 48% specificity. Conclusion Neither protein biomarkers nor machine learning algorithms are recommended alone for screening purposes because of low specificity and sensitivity values. This tool can be preferably used to improve the results of image processing methods as a complementary tool in automatic or semiautomatic systems.