Ultrasonographic Thyroid Nodule Classification with Deep Transfer Learning System Using Fine-Tuned Convolutional Neural Network
- Presentation Speakers / Moderators
Background/ Purpose: With many thyroid nodules being incidentally detected, it is important to identify as many malignant nodules as possible. The objective of this study was to develop a thyroid nodule classification system in ultrasonography using convolutional neural network.
Methods: In this study, transverse and longitudinal ultrasonographic thyroid images of 862 patients in Korea Institute of Radiological & Medical Sciences (KIRAMS) were used for training and testing a deep learning model. Of all, 325 cases were confirmed as benign, and 437 as papillary thyroid carcinoma after surgical biopsy. The images were distributed into train, validation and test sets, and ratio was set to 6:2:2. Annotation marks were removed and missing regions were recovered with neighboring parenchyme. To reduce overfitting of deep learning model, we applied data augmentation, global average pooling and 4-fold cross validation. We employed the transfer learning method with pre-trained deep learning model, VGG16. The performance of the model was evaluated using receiver operating characteristic (ROC) curve.
Results: Average AUC was 0.916, and specificity and sensitivity were 0.70 and 0.92 respectively. Positive and negative predictive value were 0.90 and 0.75.
Discussion & Conclusion: In conclusion, we introduced a new fine-tuned deep learning model for classifying thyroid nodules in USG. We expect this model will help the physicians to diagnose thyroid nodules with USG.
Two Approaches to Level 5 Neck Dissection
- Jeff Blumberg