Miss Xiuxiu Hu1, Miss Jinyue Yang2, Professor Siyu Xia2, Professor Pingsheng Chen1
1Department of Pathology, School of Medicine, Southeast University, Nanjing, China, 2School of Automation, Southeast university, Nanjing, China
Aim: To classify immunofluorescence images of kidney biopsies, a novel convolutional neural network-based approach was explored in this study.
Background: Immunofluorescence images of kidney biopsies serve as crucial roles in the diagnosis of renal pathology, since certain distinctive fluorescence details could be found in the corresponding kidney diseases. Deep learning methods have been utilized as potent tools in the classification of pathology images as well as in the disease diagnosis.
Methods: 2436 immunofluorescence images of renal biopsies from the Institute of Nephrology, Southeast University were collected. Among them, 1773 images were handled as a training set, and representative fluorescence images of membranous nephropathy, diabetic nephropathy, and IgA nephropathy were manually screened based on the fluorescence distribution characteristics combined with clinical information. A neoteric dense multi-branch convolutional network (D-ResNeXt) on basis of ResNeXt network was then proposed to sort the immunofluorescence images. The remaining 663 images were used as a test set to verify the accuracy.
Results: Uniform distributions of granular fluorescence along the glomerular capillary collaterals were revealed in membranous nephropathy, while classical linear fluorescence was distributed along the glomerular basement membrane in diabetic nephropathy, and granular or clumped fluorescence was prominently found in the glomerular thylakoid region in IgA nephropathy. Most importantly, our model performed well in the classification macroscopic accuracy, macroscopic completeness and macro F1 score, which respectively reached to 96.86%, 97.4% and 97.13%.
Conclusions: The convolutional neural network model constructed in this study provides a new strategy for efficiently and accurately classify immunofluorescence images of kidney biopsies.
Biography:
Xiuxiu Hu, female, 27 years old, graduated from Henan University of Science and Technology with a bachelor’s degree in medicine, and Southeast University with a master’s degree, and currently studying for a doctor’s degree in Southeast University. The research direction is renal pathology, and an article on this direction has been accepted. Research project is the application of artificial intelligence in the diagnosis and prognosis of kidney diseases.