Hyperparameter selection for ResNet classification of malignancy from thyroid ultrasound images

Joseph Cox; Sydney Rubin; Joe Adams; Carina Pereira; Manjiri Dighe; Adam Alessio

doi:10.1117/12.2550531

16 March 2020 Hyperparameter selection for ResNet classification of malignancy from thyroid ultrasound images

Joseph Cox, Sydney Rubin, Joe Adams, Carina Pereira, Manjiri Dighe, Adam Alessio

Proceedings Volume 11314, Medical Imaging 2020: Computer-Aided Diagnosis; 1131447 (2020) https://doi.org/10.1117/12.2550531
Event: SPIE Medical Imaging, 2020, Houston, Texas, United States

Abstract

Thyroid nodules are extremely common, with a prevalence of up to 68% in adults. Ultrasound imaging is usually performed to detect and evaluate thyroid nodules for malignancy. Many patients undergo follow-up biopsy in the form of fine-needle aspiration (FNA) to determine if a nodule is malignant or benign, although most nodules are benign. In order to reduce the number of unnecessary FNAs, radiologists will often use classification systems such as Thyroid Imaging, Reporting, and Data System (TI-RADS) to provide risk stratification and a recommendation regarding whether FNA is necessary. This scoring is both subjective and time-consuming, leading to discrepancies between radiologists and recommendations that can be inaccurate. We hypothesize that a machine learned classifier can be identified with accurate and generalizable performance, potentially offering more consistent results than manual evaluation. We created a network from two ResNet-50 branches accepting two inputs, shear-wave elastography and B-mode ultrasound images. We performed a grid search to determine the optimal hyperparameters for our model, resulting in a network that predicted malignancy of nodules with 88.7% accuracy and an AUC of 0.91. Along with identifying the training hyperparameters with optimal classification accuracy, the grid search also allowed us to select training parameters that led to more generalizable model performance on test data sets. These initial performance results suggest that our model offers a promising strategy for thyroid nodule classification and a strategy to help identify more generalizable models.

Citation Download Citation

Joseph Cox, Sydney Rubin, Joe Adams, Carina Pereira, Manjiri Dighe, and Adam Alessio "Hyperparameter selection for ResNet classification of malignancy from thyroid ultrasound images", Proc. SPIE 11314, Medical Imaging 2020: Computer-Aided Diagnosis, 1131447 (16 March 2020); https://doi.org/10.1117/12.2550531

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available

Members: $17.00

Non-members: $21.00 ADD TO CART

PROCEEDINGS
6 PAGES

DOWNLOAD PAPER SAVE TO MY LIBRARY

GET CITATION

RIGHTS & PERMISSIONS

Get copyright permission Get copyright permission on Copyright Marketplace

KEYWORDS

Data modeling

RGB color model

Ultrasonography

Performance modeling

Elastography

Image classification

Biopsy

Show All Keywords

Keywords/Phrases

Search In:

Publication Years