A hybrid descriptor to improve kidney pathologies classification
Laiara Cristina da Silva, Vinícius Ponte Machado, Rodrigo de Melo Sousa Veras, Keylla Maria de Sá Urtiga Aita, Semíramis Jamil Hadad do Monte, Nayze Lucena Sangreman Aldeman · Proceedings of the 37th ACM/SIGAPP Symposium on Applied Computing · 2022
The importance of glomerular function in kidney physiology characterizes glomerular diseases as the main problem in nephrology. So finding and classifying glomerular disorders are fundamental steps for diagnosing many kidney diseases. This paper conducted an extensive study to determine the best set of features for glomerular image representation. Our feature extraction methodology, which includes clinical data, texture, and global descriptors, resulted in 8486 features. Besides, we compared four classifiers to propose a method that helps the specialist define a renal pathology diagnosis. The proposed method achieved an accuracy of 98.46% and a Kappa index of 98.42% using the Random Forest Classifier. We concluded that a combination of clinical data and global image features facilitates accurate disease classification.