%0 Journal Article
%T Using deep convolutional neural networks for multi-classification of thyroid tumor by histopathology: a large-scale pilot study
%A Duanshu Li
%A Hongtao Lu
%A Iweng Lao
%A Jun Xiang
%A Li Wang
%A Qing Guan
%A Qinghai Ji
%A Yi Wu
%A Yongxue Zhu
%A Yu Wang
%A Yunjun Wang
%J SCIE-indexed Journal
%D 2019
%R 10.21037/atm.2019.08.54
%X Thyroid nodules are common diseases presented in the clinic, and their pathological types are complex. Thyroid nodules mainly include benign tumors and malignant tumors (i.e., thyroid cancer). Benign thyroid tumors include nodular goiter and thyroid adenoma. Thyroid cancers include papillary, follicular, medullary and anaplastic carcinomas. The differential diagnosis of thyroid nodules is crucial because thyroid carcinoma requires surgery, while only follow-up is necessary in cases of benign nodules. Pathological diagnosis of resected specimens is the gold standard for tumor diagnosis. Currently, the vast majority of pathological tissue sections are acquired by pathologists, and collections of specimens accumulated over long periods are used for clinical diagnosis. Nevertheless, manual differential diagnosis of thyroid tumor histopathological images remains a challenge for three main reasons: (I) the ability to correctly diagnose samples greatly depends on the professional background and experience of the pathologist, and such experience cannot be acquired quickly; (II) the work is tedious, expensive and time-consuming; and (III) it is challenging for the human eye to distinguish subtle changes in tissues; thus, pathologists can experience fatigue, which may lead to misdiagnosis. Thus, the precise histopathologic diagnosis of thyroid nodules remains a challenging task
%U http://atm.amegroups.com/article/view/28771/html