A Review on Deep Learning Approaches for Glomeruli Classification and Detection in Human Kidney Tissue Images

In recent years, technology has made it possible to cure millions of diseases, analyse infections, provide reliable test results, and perform high-level maintenance checks. The human body can now be diagnosed and studied at an easy level to improve health and cure illnesses. The research sector has been working at a steady pace to complete numerous studies for identification of various diseases in different body organs and to conduct a descriptive study for determination of the proper functioning mechanism of the human body. One such study, the identification of glomeruli in human kidney tissue, has likewise made significant advances in recent years. Damage to the glomeruli is a typical symptom of kidney disorders such as glomerulonephritis and glomerulosclerosis, which can progress to chronic kidney disease and kidney failure. These disorders are easily treatable if detected at an early stage [1]. Detecting and categorizing glomerular abnormalities is important in many kidney diseases. These tasks require pathologists' knowledge, and significant work has been done to define and agree on important lesions. Traditionally, biopsies of kidney tissues for glomeruli detection were evaluated visually by skilled pathologists using a light microscope which is a time-consuming and highly variable operation [2].  Although few large medical centres with in-house nephropathologists can handle this process quickly, this knowledge is not available everywhere. In the medical field, advanced techniques such as Machine Learning are increasingly being used to efficiently fulfil these tasks. Patterns and features can be correlated with clinical outcomes using Machine Learning techniques. Researchers have recently designed “Deep Learning (DL)” frameworks, such as CNN for detection and classification of different objects [3].Deep Learning has had a huge impact on the healthcare industry as technology has evolved. It is a form of structural learning technique that is trained using a hierarchical representation of data's characteristics. The advantage of deep learning is that it generates enhanced feature representations directly from the source images. By using Deep Learning, the image segmentation is used to identify the glomeruli in images and classification is used to classify the glomerular and non-glomerular images. The detection of glomeruli in different types of kidney tissue images further can help to predict the type of glomerular diseases [4]. We begin this review study by providing an overview of glomerular diseases such as glomerulonephritis and glomerulosclerosis, followed by a listing of publicly available glomeruli datasets. After that, popular deep learning model such as AlexNet, SegNet and Faster RNN are also defined.  Several performance parameters such as accuracy, precision, sensitivity, recall, F-score and dice coefficient used for the detection of glomeruli are also explained. In this review, different scientific articles based on deep learning have been thoroughly analysed and addressed for identification, segmentation and classification of glomeruli in human kidney tissue images. The brief overview of various segmentation and classification techniques in deep learning and the recent work in a conceptual and systematic manner is also discussed here. Finally, a comparative analysis of the state of art is made on the basis of the number of images, glomeruli count, technique used and type of stain used. Additionally, the difficulties encountered in the field are highlighted and recommendations are made for future study possibilities.