Рақамли технологияларнинг назарий ва амалий масалалари Том 8 № 2 (2025) · с. 37-46
Classification of Diseases Based on Kidney Computed Tomography Images
Маматов, Н.С., Джалелова, М.М., Джураев, И.А., Файзиев, В.О., Самиджонов, А.Н.
Аннотация
This research work is devoted to the problem of classifying diseases in kidney computed tomography images using neural networks. The novelty and uniqueness of the research is that it studies not only the traditional binary classification such as “normal/pathological”, but also complex differential diagnosis issues between diseases such as “cyst/stone”, “cyst/tumor”, “stone/tumor”. In the work, binary, ternary and quaternary classification models were used to identify kidney conditions such as normal, cyst, stone, tumor. Depending on the level of complexity of the task, i.e., for binary and ternary classification, architectures such as MobileNetV2 were used, and for quaternary classification, architectures such as VGG16 were used. According to the results, all models achieved high accuracy, with F1-scores of 99.39-100% for the stone class, 98.99-100% for the tumor class, and 99.60% for the cyst/tumor pair. The proposed combinatorial approach, i.e., initially making a general diagnosis using the four-class classification model, and in doubtful cases, conducting additional examination with binary or ternary models, is a reasonable approach for clinical practice. The results of the study will serve as a valuable tool for radiologists and urologists to accurately diagnose kidney diseases and choose the optimal treatment strategy, in accordance with their actual clinical needs.
заболевания почексверточная нейронная сетькомпьютерно-томографическое изображениеточностьполнотаf1-оценкабинарная классификациямногоклассовая классификацияkidney diseasesconvolutional neural network
Источник метаданных: OAI-PMH архив журнала · Sindex не хранит полный текст, а даёт ссылку на источник.