| Abstract [eng] |
Background. The histological diagnosis of glomerular diseases requires manual lesion classification and is a crucial step in clinical decision-making. This complex and time-consuming task often yields inconsistent results among experienced pathologists. Innovations in artificial intelligence hold potential to resolve these issues. Objective. To investigate the progress of the latest artificial intelligence-based methods and summarize the research results in this field, discuss challenges and future directions in the investigation and implementation of artificial intelligence methods in pathological diagnosis of glomerular lesions. Methods. Literature review was performed in PubMed and Google Scholar databases, using combinations of these keywords: ("machine learning" or "artificial intelligence" or "convolutional neural network" or "CNN" or "deep learning" or "computational pathology" or "digital pathology") and ("renal pathology" or "nephropathology" or "glomerul*" or "kidney"). Results. The deployment of artificial intelligence methodologies has yielded encouraging outcomes in detecting and segmenting glomeruli and further distinguishing between healthy and damaged structures in whole slide images. High accuracy has been attained in the automation of glomerular cell proliferation analysis and the categorization of glomeruli into normal and sclerotic. Ongoing research endeavors are focused on a more routine diagnostic process, which includes evaluation of multiple lesions in a single glomerulus, interpretation of immunofluorescence images or other pertinent supplementary investigations, and correlation of pathological and clinical indicators to ascertain patient prognosis. Conclusions. The application of automation in the routine diagnostic evaluation of glomeruli is infrequent due to the diversity of histological presentations of glomerular diseases, the complexity inherent in diagnosis, variability in tissue processing methods across laboratories, data deficiency, validation challenges, and legal constraints. Implementing segmentation of glomerular structures and lesions could potentially enhance algorithmic accuracy. Future initiatives should include the development of digital repositories with standardized protocols for scanning, uploading, and annotating pathological material; acceleration of image annotation techniques; and development of methodologies to simplify the interpretation of automated analysis results. |
