Classification of glomerular hypercellularity using convolutional\n features and support vector machine

Glomeruli are histological structures of the kidney cortex formed by\ninterwoven blood capillaries, and are responsible for blood filtration.\nGlomerular lesions impair kidney filtration capability, leading to protein loss\nand metabolic waste retention. An example of lesion is the glomerular\nhypercellularity, which is characterized by an increase in the number of cell\nnuclei in different areas of the glomeruli. Glomerular hypercellularity is a\nfrequent lesion present in different kidney diseases. Automatic detection of\nglomerular hypercellularity would accelerate the screening of scanned\nhistological slides for the lesion, enhancing clinical diagnosis. Having this\nin mind, we propose a new approach for classification of hypercellularity in\nhuman kidney images. Our proposed method introduces a novel architecture of a\nconvolutional neural network (CNN) along with a support vector machine,\nachieving near perfect average results with the FIOCRUZ data set in a binary\nclassification (lesion or normal). Our deep-based classifier outperformed the\nstate-of-the-art results on the same data set. Additionally, classification of\nhypercellularity sub-lesions was also performed, considering mesangial,\nendocapilar and both lesions; in this multi-classification task, our proposed\nmethod just failed in 4\\% of the cases. To the best of our knowledge, this is\nthe first study on deep learning over a data set of glomerular hypercellularity\nimages of human kidney.\n