[Prospects for the application of convolutional neural networks in the cytological diagnosis of thyroid nodules].

Aim:  Analysis and assessment of the role of convolutional neural networks in the cytological diagnosis of the thyroid pathology, exploring their potential for increasing the accuracy and automation of diagnostic processes.

Methods:  Analysis of literature from Pubmed, Google Scholar and the scientific electronic library elibrary.ru using the keywords «thyroid», «cytology», «cytopathology», «fine-needle aspiration biopsy», «neural network» and «convolutional neural network». 12 articles published from 2018 to 2023 were selected for analysis.

Results:  The paper discusses the basic principles of the design of convolutional neural networks and the metrics that are used to assess their quality. An analysis of studies on the use of convolutional neural networks in the cytological diagnosis of the thyroid pathology was performed. According to the results, these neural networks classify pathological conditions with high accuracy and sensitivity, comparable to the work of an experienced cytologist. The accuracy of classification of papillary carcinoma can reach 99.7%. However, the lack of uniform standards for preparing images for training neural networks, the insufficient number of studies using multicenter data, and the narrow diagnostic range of available neural network models still limit the implementation of such AI systems in cytological diagnostic practice.

Conclusion:  The available research results on various options for using convolutional neural networks in the cytological diagnosis of the thyroid pathology have every chance of becoming the initiator of a serious paradigm shift in conventional cytopathology towards digital and computational cytopathology, in which the main functions will be performed by AI systems.