FOR RESEARCH USE ONLY
This classification is intended to academic evaluation of the potentiality of CNN in Renal Pathology.
NephroNN is an automated tool for immunofluorescence classification, to support the diagnosis process of renal biopsies. Thanks to the web application designed by the AImageLab research group in Collaboration with the Division of Nephrology Dialysis and Transplantation of the AOU Policlinico di Modena, pathologists all around the world can now assess neural networks performances, by uploading their own images to obtain predictions based on Deep Learning! You can sign up right now and upload an image, NephroNN will evaluate Appearance, Distribution, Location and Intensity of the deposits according to 9 different features:
Location | Mesangial | 1 | |
Parietal | 2 | ||
Continuous Regular Parietal | 3 | ||
Irregular Parietal | 4 | ||
Appearance | Coarse Granular | 5 | |
Fine Granular | 6 | ||
Distribution | Segmental | 7 | |
Global | 8 | ||
Intensity | 9 |
A Deep Convolutional Network was trained for each different task, taking advantage of the dataset collected at the University of Modena and Reggio Emilia. Every network learns to recognize the features as they were labeled by expert pathologists and to generalize its pattern recognition abilities in order to analyze new images. More details are available in the "paper under review". You can check out the images that were made public by visiting the Public Images section of this website, every image was automatically transformed into grayscale by extracting the green channel when uploaded. All you have to do now is sign up and upload an image:
the immunofluorescence image should be captured at X400 magnification. Multiple formats are supported, check out the complete list here. You can upload either RGB/grayscale images or 16 bits per pixel tiff images.
please anonymize the images before uploading! The University of Modena and Reggio Emilia will not be liable in any way for damages caused by the exposure of sensitive data on the NephroNN server because of the missing anonymization of clinical images.