(Wednesday, 7/17/2019) 4:30 PM - 6:00 PM
Room: 303
Purpose: Thyroid Imaging Reporting and Data System (TI-RADS) class-4 is considered as malignant thyroid nodule with suspicious ultrasound features. Ultrasound-guided needle aspiration biopsy is the most effective method for the diagnosis of class-4 thyroid nodules. However, needle aspiration biopsy for all the patients with thyroid nodules is expensive and unrealistic, accurate and cost-effective classification of class-4 thyroid nodule is needed.
Methods: A multi-layer perception based classifier was proposed for TI-RADS class-4 thyroid nodules classification. Patients with thyroid nodules were collected and received biopsy examination. Patients with TI-RADS class 4a, 4b and 4c were classified and relevant ultrasound images were obtained. A 128*128-pixel image containing thyroid nodule was extracted from each of the original ultrasound image in order to remove annotations and markers. To reduce computational complexity, the 128*128-pixel images were resized into 32*32 images. Then data normalization was performed. The dataset was separated into training set and testing set. Labels with -1 for TI-RADS 4a patients, 0 for TI-RADS 4b patients and 1 for TI-RADS 4c patients were used. A multi-layer perception consisted of two layers was configured with 9 neurons in the input layer and 1 neuron in the output layer. Tanh function was defined as activation function for the input and output layer. Mean-square error was defined as cost function and Levenberg-Marquardt algorithm was used as update algorithm for the multi-layer perception. The multi-layer perception was trained in the training set and performance was evaluated in the testing set.
Results: The results demonstrated that the proposed multi-layer perception can accurately and effectively diagnose TI-RADS class-4 thyroid nodules with a total accuracy of 91.67%. The classification accuracies for TI-RADS 4a, 4b and 4c were 100%, 100% and 75% respectively.
Conclusion: The multi-layer perception based classifier is cost-effective and the results demonstrates its potential clinical application in thyroid nodules diagnosis.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by National Key Research and Development Program of China (2016YFC0103400), Key Research and Development Program of Shandong Province (2017GGX201010), Jianfeng Q. was supported by the Taishan Scholars Program of Shandong Province (TS201712065).
Keywords
Ultrasonics, Image Analysis, Classifier Design
Taxonomy
IM- Ultrasound : Machine learning, computer vision
Contact Email
