Title: Deep convolutional neural network models for the diagnosis of thyroid cancer
Abstract: The study by Xiangchun Li and colleagues 1 Li X Zhang S Zhang Q et al. Diagnosis of thyroid cancer using deep convolutional neural network models applied to sonographic images: a retrospective, multicohort, diagnostic study. Lancet Oncol. 2019; 20: 193-201 Summary Full Text Full Text PDF PubMed Scopus (187) Google Scholar adds to the growing body of evidence that application of the newly developed deep convolutional neural network models on sonographic images can improve accuracy, sensitivity, and specificity in identifying patients with thyroid cancer at levels similar to or higher than skilled radiologists. The deep convolutional neural network model is a key component of the deep learning framework, and it has been widely employed to analyse visual imagery. 2 Khosravi P Kazemi E Imielinski M Elemento O Hajirasouliha I Deep convolutional neural networks enable discrimination of heterogeneous digital pathology images. EBioMedicine. 2018; 27: 317-328 Summary Full Text Full Text PDF PubMed Scopus (168) Google Scholar , 3 Weichenthal S Hatzopoulou M Brauer M A picture tells a thousand…exposures: opportunities and challenges of deep learning image analyses in exposure science and environmental epidemiology. Environ Int. 2019; 122: 3-10 Crossref PubMed Scopus (57) Google Scholar We agree that a reliable deep learning model can broadly influence clinical practice. Li and colleagues 1 Li X Zhang S Zhang Q et al. Diagnosis of thyroid cancer using deep convolutional neural network models applied to sonographic images: a retrospective, multicohort, diagnostic study. Lancet Oncol. 2019; 20: 193-201 Summary Full Text Full Text PDF PubMed Scopus (187) Google Scholar developed and validated the deep convolutional neural network algorithms using the largest number of images to date, yet the accuracy in three small-scale validation sets is not satisfactory, ranging from 0·857 to 0·889. Apart from sample size, several factors could be considered when interpreting the findings by Li and colleagues. 1 Li X Zhang S Zhang Q et al. Diagnosis of thyroid cancer using deep convolutional neural network models applied to sonographic images: a retrospective, multicohort, diagnostic study. Lancet Oncol. 2019; 20: 193-201 Summary Full Text Full Text PDF PubMed Scopus (187) Google Scholar In this study, the deep convolutional neural network models were merely developed on the basis of sonographic images. Given that thyroid cancer is a complex and heterogeneous disease, 4 Kasaian K Wiseman SM Walker BA et al. The genomic and transcriptomic landscape of anaplastic thyroid cancer: implications for therapy. BMC Cancer. 2015; 15: 984 Crossref PubMed Scopus (45) Google Scholar data from multiple sources, such as demographic characteristics, laboratory test results, and images, are available in practical clinical scenarios. To create a reliable diagnostic model, a more robust deep learning model combining different types of medical data sources is encouraged, and it is technically feasible at present. 5 Miotto R Wang F Wang S Jiang X Dudley JT Deep learning for healthcare: review, opportunities and challenges. Brief Bioinform. 2018; 19: 1236-1246 Crossref PubMed Scopus (1029) Google Scholar The image-based deep convolutional neural network model developed by Li and colleagues 4 Kasaian K Wiseman SM Walker BA et al. The genomic and transcriptomic landscape of anaplastic thyroid cancer: implications for therapy. BMC Cancer. 2015; 15: 984 Crossref PubMed Scopus (45) Google Scholar will be more generalisable and less susceptible to bias pending further, more comprehensive explorations. Deep convolutional neural network models for the diagnosis of thyroid cancer – Authors' replyWe appreciate the comments from Dan Hu and colleagues and Eun Ha and colleagues about our Article.1 We agree with Hu and colleagues regarding the incorporation of demographic features and laboratory test results in the model. Specifically, two neural networks can be devised that tailor for demographic features and laboratory test results. Subsequently, these two networks can be combined into a new model with the existing deep convolutional neural network model. Data curation from clinical and laboratory reports is ongoing. Full-Text PDF Diagnosis of thyroid cancer using deep convolutional neural network models applied to sonographic images: a retrospective, multicohort, diagnostic studyThe DCNN model showed similar sensitivity and improved specificity in identifying patients with thyroid cancer compared with a group of skilled radiologists. The improved technical performance of the DCNN model warrants further investigation as part of randomised clinical trials. Full-Text PDF Deep convolutional neural network models for the diagnosis of thyroid cancerWe read the paper by Xiangchun Li and colleagues,1 in which the authors describe a newly developed deep convolutional neural network model that can achieve high accuracy, sensitivity, and specificity in automated thyroid cancer diagnosis in a real-world setting. Full-Text PDF