Using Deep Convolutional Neural Networks For Multi-Classification Of Thyroid Tumor In Histopathology
- Presentation Speakers / Moderators
PURPOSE?In this study, we exploited multiple deep convolutional neural network (DCNN) models in differential diagnoses of thyroid nodule histopathological images.
METHODS: Our method covers 7 common types of thyroid diseases (i.e., normal tissue, adenoma, nodular goiter, papillary thyroid carcinoma (PTC), follicular thyroid carcinoma (FTC), medullary thyroid carcinoma (MTC) and anaplastic thyroid carcinoma (ATC)). We developed 2 DCNN models by leveraging a dataset of 9795 crop images, which were annotated by experienced expert pathologists as reference standards. We evaluated our methods on an independent test set of 1919 crop images and measured their diagnostic accuracy rates.
RESULTS: 632 hematoxylin-eosin (H&E) staining histopathologic whole-slide images of thyroid nodules were obtained to build the dataset. In the test group, VGG-19 model shows a better performance for classification of thyroid nodules than Inception_resnet_v2 model (accuracy in normal tissue: 87.4% vs. 80.0%; nodular goiter: 100% vs. 98.2%; adenoma: 92.6% vs. 90.3%; PTC: 98.2% vs. 93.0%; FTC: 98.3% vs. 94.5%; MTC: 100% vs. 98.2%; ATC: 98.3% vs. 93.89%; overall accuracy: 97.0% vs. 93.8%).
CONCLUSION: In summary, after training with a large dataset, the DCNN VGG-19 model showed great potential in facilitating classifications of thyroid diagnosis from histological images. Our method may help to reduce the cognitive burden on pathologists and improve their efficacy in histopathological characterization of thyroid nodules.
Two Approaches to Level 5 Neck Dissection
- Jeff Blumberg