Reduction of unnecessary thyroid biopsies using deep learning

Zeynettin Akkus, Arunnit Boonrod, Mahfuzur R. Siddiquee, Kenneth A. Philbrick, Marius N. Stan, M. Regina Castro, Dana Erickson, Matthew R Callstrom, Bradley J Erickson

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Thyroid nodules are extremely common lesions and highly detectable by ultrasound (US). Several studies have shown that the overall incidence of papillary thyroid cancer in patients with nodules selected for biopsy is only about 10%. Therefore, there is a clinical need for a dramatic reduction of thyroid biopsies. In this study, we present a guided classification system using deep learning that predicts malignancy of nodules from B-mode US. We retrospectively collected transverse and longitudinal images of 150 benign and 150 malignant thyroid nodules with biopsy proven results. We divided our dataset into training (n=460), validation(n=40), and test (n=100) datasets. We manually segmented nodules from B-mode US images and provided the nodule mask as a second input channel to the convolutional neural network (CNN) for increasing the attention of nodule regions in images. We evaluated the classification performance of different CNN architectures such as Inception and Resnet50 CNN architectures with different input images. The InceptionV3 model showed the best performance on the test dataset: 86% (sensitivity), 90% (specificity), and 90% precision when the threshold was set for highest accuracy. When the threshold was set for maximum sensitivity (0 missed cancers), the ROC curve suggests the number of biopsies may be reduced by 52% without missing patients with malignant thyroid nodules. We anticipate that this performance can be further improved with including more patients and the information from other ultrasound modalities.

Original languageEnglish (US)
Title of host publicationMedical Imaging 2019
Subtitle of host publicationImage Processing
EditorsBennett A. Landman, Elsa D. Angelini, Elsa D. Angelini, Elsa D. Angelini
PublisherSPIE
ISBN (Electronic)9781510625457
DOIs
StatePublished - Jan 1 2019
EventMedical Imaging 2019: Image Processing - San Diego, United States
Duration: Feb 19 2019Feb 21 2019

Publication series

NameProgress in Biomedical Optics and Imaging - Proceedings of SPIE
Volume10949
ISSN (Print)1605-7422

Conference

ConferenceMedical Imaging 2019: Image Processing
CountryUnited States
CitySan Diego
Period2/19/192/21/19

Fingerprint

Biopsy
nodules
learning
Thyroid Nodule
Thyroid Gland
Ultrasonics
Learning
Neural networks
Network architecture
Masks
ROC Curve
Neoplasms
cancer
Sensitivity and Specificity
thresholds
Deep learning
sensitivity
Incidence
lesions
Datasets

Keywords

  • benign thyroid nodule
  • convolutional neural networks
  • malignant thyroid nodule.
  • nodule classification
  • Thyroid cancer
  • thyroid nodules
  • thyroid ultrasound

ASJC Scopus subject areas

  • Electronic, Optical and Magnetic Materials
  • Atomic and Molecular Physics, and Optics
  • Biomaterials
  • Radiology Nuclear Medicine and imaging

Cite this

Akkus, Z., Boonrod, A., Siddiquee, M. R., Philbrick, K. A., Stan, M. N., Castro, M. R., ... Erickson, B. J. (2019). Reduction of unnecessary thyroid biopsies using deep learning. In B. A. Landman, E. D. Angelini, E. D. Angelini, & E. D. Angelini (Eds.), Medical Imaging 2019: Image Processing [109490W] (Progress in Biomedical Optics and Imaging - Proceedings of SPIE; Vol. 10949). SPIE. https://doi.org/10.1117/12.2512574

Reduction of unnecessary thyroid biopsies using deep learning. / Akkus, Zeynettin; Boonrod, Arunnit; Siddiquee, Mahfuzur R.; Philbrick, Kenneth A.; Stan, Marius N.; Castro, M. Regina; Erickson, Dana; Callstrom, Matthew R; Erickson, Bradley J.

Medical Imaging 2019: Image Processing. ed. / Bennett A. Landman; Elsa D. Angelini; Elsa D. Angelini; Elsa D. Angelini. SPIE, 2019. 109490W (Progress in Biomedical Optics and Imaging - Proceedings of SPIE; Vol. 10949).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Akkus, Z, Boonrod, A, Siddiquee, MR, Philbrick, KA, Stan, MN, Castro, MR, Erickson, D, Callstrom, MR & Erickson, BJ 2019, Reduction of unnecessary thyroid biopsies using deep learning. in BA Landman, ED Angelini, ED Angelini & ED Angelini (eds), Medical Imaging 2019: Image Processing., 109490W, Progress in Biomedical Optics and Imaging - Proceedings of SPIE, vol. 10949, SPIE, Medical Imaging 2019: Image Processing, San Diego, United States, 2/19/19. https://doi.org/10.1117/12.2512574
Akkus Z, Boonrod A, Siddiquee MR, Philbrick KA, Stan MN, Castro MR et al. Reduction of unnecessary thyroid biopsies using deep learning. In Landman BA, Angelini ED, Angelini ED, Angelini ED, editors, Medical Imaging 2019: Image Processing. SPIE. 2019. 109490W. (Progress in Biomedical Optics and Imaging - Proceedings of SPIE). https://doi.org/10.1117/12.2512574
Akkus, Zeynettin ; Boonrod, Arunnit ; Siddiquee, Mahfuzur R. ; Philbrick, Kenneth A. ; Stan, Marius N. ; Castro, M. Regina ; Erickson, Dana ; Callstrom, Matthew R ; Erickson, Bradley J. / Reduction of unnecessary thyroid biopsies using deep learning. Medical Imaging 2019: Image Processing. editor / Bennett A. Landman ; Elsa D. Angelini ; Elsa D. Angelini ; Elsa D. Angelini. SPIE, 2019. (Progress in Biomedical Optics and Imaging - Proceedings of SPIE).
@inproceedings{8e248695d42747aa84337267cb8cf8ab,
title = "Reduction of unnecessary thyroid biopsies using deep learning",
abstract = "Thyroid nodules are extremely common lesions and highly detectable by ultrasound (US). Several studies have shown that the overall incidence of papillary thyroid cancer in patients with nodules selected for biopsy is only about 10{\%}. Therefore, there is a clinical need for a dramatic reduction of thyroid biopsies. In this study, we present a guided classification system using deep learning that predicts malignancy of nodules from B-mode US. We retrospectively collected transverse and longitudinal images of 150 benign and 150 malignant thyroid nodules with biopsy proven results. We divided our dataset into training (n=460), validation(n=40), and test (n=100) datasets. We manually segmented nodules from B-mode US images and provided the nodule mask as a second input channel to the convolutional neural network (CNN) for increasing the attention of nodule regions in images. We evaluated the classification performance of different CNN architectures such as Inception and Resnet50 CNN architectures with different input images. The InceptionV3 model showed the best performance on the test dataset: 86{\%} (sensitivity), 90{\%} (specificity), and 90{\%} precision when the threshold was set for highest accuracy. When the threshold was set for maximum sensitivity (0 missed cancers), the ROC curve suggests the number of biopsies may be reduced by 52{\%} without missing patients with malignant thyroid nodules. We anticipate that this performance can be further improved with including more patients and the information from other ultrasound modalities.",
keywords = "benign thyroid nodule, convolutional neural networks, malignant thyroid nodule., nodule classification, Thyroid cancer, thyroid nodules, thyroid ultrasound",
author = "Zeynettin Akkus and Arunnit Boonrod and Siddiquee, {Mahfuzur R.} and Philbrick, {Kenneth A.} and Stan, {Marius N.} and Castro, {M. Regina} and Dana Erickson and Callstrom, {Matthew R} and Erickson, {Bradley J}",
year = "2019",
month = "1",
day = "1",
doi = "10.1117/12.2512574",
language = "English (US)",
series = "Progress in Biomedical Optics and Imaging - Proceedings of SPIE",
publisher = "SPIE",
editor = "Landman, {Bennett A.} and Angelini, {Elsa D.} and Angelini, {Elsa D.} and Angelini, {Elsa D.}",
booktitle = "Medical Imaging 2019",

}

TY - GEN

T1 - Reduction of unnecessary thyroid biopsies using deep learning

AU - Akkus, Zeynettin

AU - Boonrod, Arunnit

AU - Siddiquee, Mahfuzur R.

AU - Philbrick, Kenneth A.

AU - Stan, Marius N.

AU - Castro, M. Regina

AU - Erickson, Dana

AU - Callstrom, Matthew R

AU - Erickson, Bradley J

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Thyroid nodules are extremely common lesions and highly detectable by ultrasound (US). Several studies have shown that the overall incidence of papillary thyroid cancer in patients with nodules selected for biopsy is only about 10%. Therefore, there is a clinical need for a dramatic reduction of thyroid biopsies. In this study, we present a guided classification system using deep learning that predicts malignancy of nodules from B-mode US. We retrospectively collected transverse and longitudinal images of 150 benign and 150 malignant thyroid nodules with biopsy proven results. We divided our dataset into training (n=460), validation(n=40), and test (n=100) datasets. We manually segmented nodules from B-mode US images and provided the nodule mask as a second input channel to the convolutional neural network (CNN) for increasing the attention of nodule regions in images. We evaluated the classification performance of different CNN architectures such as Inception and Resnet50 CNN architectures with different input images. The InceptionV3 model showed the best performance on the test dataset: 86% (sensitivity), 90% (specificity), and 90% precision when the threshold was set for highest accuracy. When the threshold was set for maximum sensitivity (0 missed cancers), the ROC curve suggests the number of biopsies may be reduced by 52% without missing patients with malignant thyroid nodules. We anticipate that this performance can be further improved with including more patients and the information from other ultrasound modalities.

AB - Thyroid nodules are extremely common lesions and highly detectable by ultrasound (US). Several studies have shown that the overall incidence of papillary thyroid cancer in patients with nodules selected for biopsy is only about 10%. Therefore, there is a clinical need for a dramatic reduction of thyroid biopsies. In this study, we present a guided classification system using deep learning that predicts malignancy of nodules from B-mode US. We retrospectively collected transverse and longitudinal images of 150 benign and 150 malignant thyroid nodules with biopsy proven results. We divided our dataset into training (n=460), validation(n=40), and test (n=100) datasets. We manually segmented nodules from B-mode US images and provided the nodule mask as a second input channel to the convolutional neural network (CNN) for increasing the attention of nodule regions in images. We evaluated the classification performance of different CNN architectures such as Inception and Resnet50 CNN architectures with different input images. The InceptionV3 model showed the best performance on the test dataset: 86% (sensitivity), 90% (specificity), and 90% precision when the threshold was set for highest accuracy. When the threshold was set for maximum sensitivity (0 missed cancers), the ROC curve suggests the number of biopsies may be reduced by 52% without missing patients with malignant thyroid nodules. We anticipate that this performance can be further improved with including more patients and the information from other ultrasound modalities.

KW - benign thyroid nodule

KW - convolutional neural networks

KW - malignant thyroid nodule.

KW - nodule classification

KW - Thyroid cancer

KW - thyroid nodules

KW - thyroid ultrasound

UR - http://www.scopus.com/inward/record.url?scp=85068351560&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85068351560&partnerID=8YFLogxK

U2 - 10.1117/12.2512574

DO - 10.1117/12.2512574

M3 - Conference contribution

AN - SCOPUS:85068351560

T3 - Progress in Biomedical Optics and Imaging - Proceedings of SPIE

BT - Medical Imaging 2019

A2 - Landman, Bennett A.

A2 - Angelini, Elsa D.

A2 - Angelini, Elsa D.

A2 - Angelini, Elsa D.

PB - SPIE

ER -