A review on various methods for recognition of urine particles using digital microscopic images of urine sediments. (July 2021)

• Record Type: Journal Article
• Authors: K., Suhail; D., Brindha
• Is Part Of: Biomedical signal processing and control. Volume 68(2021)
• Publication Date: 2021-07
• Subjects: CNN -- RCNN -- SSD -- LeNet-5; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102806
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Automated urinalysis based on conventional machine learning methods. Deep learning based urine sediment images analysis. Diagnosing kidney diseases by urine particles recognition. Abstract: Urine sediment examination is important for any patient with renal disease. Urinalysis may be physical, chemical or microscopic examinations. Microscopic examination determines the parameters such as Red Blood Cells (RBC), White Blood Cells (WBC), Epithelial Cells, Crystals, Bacteria, and Casts. Results from this test identify various kidney-related diseases such as Hematuria, Kidney Stones, etc. This review compares various automated methods used for urinalysis. The traditional method for microscopic examination of urine sediment performed manually from centrifuged urine samples. It is a time-consuming process and there is possibility of manual errors. This work describes the classification of microscopic images of urine sediments by conventional automated microscopic techniques and by using different types of convolutional neural networks (CNN). The problem with the conventional automated models is that the segmentation and feature extraction to be carefully designed. The characteristics of microscopic urine images make it a formidable task. The convolutional neural network classifies the images without feature extraction and segmentation. Various convolutional neural networks proposed in the literature are different types of RCNN, SSD and its variants and LeNet-5 neural network.