Classification of diabetes maculopathy images using data-adaptive neuro-fuzzy inference classifier
Prolonged diabetes retinopathy leads to diabetes maculopathy, which causes gradual and irreversible loss of vision. It is important for physicians to have a decision system that detects the early symptoms of the disease. This can be achieved by building a classification model using machine learning...
Saved in:
Published in | Medical & biological engineering & computing Vol. 53; no. 12; pp. 1345 - 1360 |
---|---|
Main Authors | , , , , , , |
Format | Journal Article |
Language | English |
Published |
Berlin/Heidelberg
Springer Berlin Heidelberg
01.12.2015
Springer Nature B.V |
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | Prolonged diabetes retinopathy leads to diabetes maculopathy, which causes gradual and irreversible loss of vision. It is important for physicians to have a decision system that detects the early symptoms of the disease. This can be achieved by building a classification model using machine learning algorithms. Fuzzy logic classifiers group data elements with a degree of membership in multiple classes by defining membership functions for each attribute. Various methods have been proposed to determine the partitioning of membership functions in a fuzzy logic inference system. A
clustering method
partitions the membership functions by grouping data that have high similarity into clusters, while an
equalized universe method
partitions data into predefined equal clusters. The distribution of each attribute determines its partitioning as fine or coarse. A simple grid partitioning partitions each attribute equally and is therefore not effective in handling varying distribution amongst the attributes. A data-adaptive method uses a data frequency-driven approach to partition each attribute based on the distribution of data in that attribute. A data-adaptive neuro-fuzzy inference system creates corresponding rules for both finely distributed and coarsely distributed attributes. This method produced more useful rules and a more effective classification system. We obtained an overall accuracy of 98.55 %. |
---|---|
Bibliography: | ObjectType-Article-1 SourceType-Scholarly Journals-1 ObjectType-Feature-2 content type line 23 |
ISSN: | 0140-0118 1741-0444 |
DOI: | 10.1007/s11517-015-1329-0 |