The image-based analysis and classification of urine sediments using a LeNet-5 neural network

Abstract

In this work, we presented a deep learning approach based on the LeNet-5 network for analysing and classifying recognisable shapes in urine sample images. The approach is based on shape analysis to recognise and classify red blood cells, white blood cells, epithelial cells and crystals observed under microscopes in urine samples. We modified the LeNet-5 neural network by changing the numbers of output nodes and convolutional layers. We compared the results of our method with those obtained by traditional feature extraction followed by back-propagation neural networks. Our testing showed that our method achieved a higher accuracy, sensitivity and specificity. The performance of our method demonstrated its broad applicability in urine sample analysis.