Comparison of deep learning and human observer performance for lesion detection and characterization

Ruben De Man, Grace J. Gang, Xin Li, Ge Wang

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

Abstract

The detection and characterizations of abnormalities in clinical imaging is of the utmost importance for patient diagnosis and treatment. In this paper, we present a comparison of convolutional neural network (CNN) and human observer performance on a simulated lesion detection and characterization task. We apply both conventional performance metrics including accuracy and non-conventional metrics such as lift charts to perform qualitative and quantitative comparison of each type of observer. It is determined that the CNN generally outperforms the human observers, particularly at high noise levels. However, high noise correlation reduces the relative performance of the CNN, and human observer performance is comparable to CNN under these conditions. These findings extend into the field of diagnostic radiology, where the adoption of deep learning is starting to become widespread. The importance of considering the applications for which deep learning is most effective is of critical importance to this development.

Original languageEnglish (US)
Title of host publication15th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine
EditorsSamuel Matej, Scott D. Metzler
PublisherSPIE
ISBN (Electronic)9781510628373
DOIs
StatePublished - Jan 1 2019
Event15th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine, Fully3D 2019 - Philadelphia, United States
Duration: Jun 2 2019Jun 6 2019

Publication series

NameProceedings of SPIE - The International Society for Optical Engineering
Volume11072
ISSN (Print)0277-786X
ISSN (Electronic)1996-756X

Conference

Conference15th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine, Fully3D 2019
CountryUnited States
CityPhiladelphia
Period6/2/196/6/19

Fingerprint

lesions
learning
Observer
Neural Networks
Neural networks
Radiology
radiology
Medical imaging
abnormalities
charts
Performance Metrics
Chart
Diagnostics
Imaging
Metric
Human
Learning
Deep learning

Keywords

  • Artificial intelligence
  • Detection
  • Image analysis
  • Image quality
  • Noise

ASJC Scopus subject areas

  • Electronic, Optical and Magnetic Materials
  • Condensed Matter Physics
  • Computer Science Applications
  • Applied Mathematics
  • Electrical and Electronic Engineering

Cite this

De Man, R., Gang, G. J., Li, X., & Wang, G. (2019). Comparison of deep learning and human observer performance for lesion detection and characterization. In S. Matej, & S. D. Metzler (Eds.), 15th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine [110721F] (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 11072). SPIE. https://doi.org/10.1117/12.2532331

Comparison of deep learning and human observer performance for lesion detection and characterization. / De Man, Ruben; Gang, Grace J.; Li, Xin; Wang, Ge.

15th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine. ed. / Samuel Matej; Scott D. Metzler. SPIE, 2019. 110721F (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 11072).

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

De Man, R, Gang, GJ, Li, X & Wang, G 2019, Comparison of deep learning and human observer performance for lesion detection and characterization. in S Matej & SD Metzler (eds), 15th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine., 110721F, Proceedings of SPIE - The International Society for Optical Engineering, vol. 11072, SPIE, 15th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine, Fully3D 2019, Philadelphia, United States, 6/2/19. https://doi.org/10.1117/12.2532331
De Man R, Gang GJ, Li X, Wang G. Comparison of deep learning and human observer performance for lesion detection and characterization. In Matej S, Metzler SD, editors, 15th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine. SPIE. 2019. 110721F. (Proceedings of SPIE - The International Society for Optical Engineering). https://doi.org/10.1117/12.2532331
De Man, Ruben ; Gang, Grace J. ; Li, Xin ; Wang, Ge. / Comparison of deep learning and human observer performance for lesion detection and characterization. 15th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine. editor / Samuel Matej ; Scott D. Metzler. SPIE, 2019. (Proceedings of SPIE - The International Society for Optical Engineering).
@inproceedings{405cfc9a6b184305b4db9679228444e2,
title = "Comparison of deep learning and human observer performance for lesion detection and characterization",
abstract = "The detection and characterizations of abnormalities in clinical imaging is of the utmost importance for patient diagnosis and treatment. In this paper, we present a comparison of convolutional neural network (CNN) and human observer performance on a simulated lesion detection and characterization task. We apply both conventional performance metrics including accuracy and non-conventional metrics such as lift charts to perform qualitative and quantitative comparison of each type of observer. It is determined that the CNN generally outperforms the human observers, particularly at high noise levels. However, high noise correlation reduces the relative performance of the CNN, and human observer performance is comparable to CNN under these conditions. These findings extend into the field of diagnostic radiology, where the adoption of deep learning is starting to become widespread. The importance of considering the applications for which deep learning is most effective is of critical importance to this development.",
keywords = "Artificial intelligence, Detection, Image analysis, Image quality, Noise",
author = "{De Man}, Ruben and Gang, {Grace J.} and Xin Li and Ge Wang",
year = "2019",
month = "1",
day = "1",
doi = "10.1117/12.2532331",
language = "English (US)",
series = "Proceedings of SPIE - The International Society for Optical Engineering",
publisher = "SPIE",
editor = "Samuel Matej and Metzler, {Scott D.}",
booktitle = "15th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine",

}

TY - GEN

T1 - Comparison of deep learning and human observer performance for lesion detection and characterization

AU - De Man, Ruben

AU - Gang, Grace J.

AU - Li, Xin

AU - Wang, Ge

PY - 2019/1/1

Y1 - 2019/1/1

N2 - The detection and characterizations of abnormalities in clinical imaging is of the utmost importance for patient diagnosis and treatment. In this paper, we present a comparison of convolutional neural network (CNN) and human observer performance on a simulated lesion detection and characterization task. We apply both conventional performance metrics including accuracy and non-conventional metrics such as lift charts to perform qualitative and quantitative comparison of each type of observer. It is determined that the CNN generally outperforms the human observers, particularly at high noise levels. However, high noise correlation reduces the relative performance of the CNN, and human observer performance is comparable to CNN under these conditions. These findings extend into the field of diagnostic radiology, where the adoption of deep learning is starting to become widespread. The importance of considering the applications for which deep learning is most effective is of critical importance to this development.

AB - The detection and characterizations of abnormalities in clinical imaging is of the utmost importance for patient diagnosis and treatment. In this paper, we present a comparison of convolutional neural network (CNN) and human observer performance on a simulated lesion detection and characterization task. We apply both conventional performance metrics including accuracy and non-conventional metrics such as lift charts to perform qualitative and quantitative comparison of each type of observer. It is determined that the CNN generally outperforms the human observers, particularly at high noise levels. However, high noise correlation reduces the relative performance of the CNN, and human observer performance is comparable to CNN under these conditions. These findings extend into the field of diagnostic radiology, where the adoption of deep learning is starting to become widespread. The importance of considering the applications for which deep learning is most effective is of critical importance to this development.

KW - Artificial intelligence

KW - Detection

KW - Image analysis

KW - Image quality

KW - Noise

UR - http://www.scopus.com/inward/record.url?scp=85074252570&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85074252570&partnerID=8YFLogxK

U2 - 10.1117/12.2532331

DO - 10.1117/12.2532331

M3 - Conference contribution

AN - SCOPUS:85074252570

T3 - Proceedings of SPIE - The International Society for Optical Engineering

BT - 15th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine

A2 - Matej, Samuel

A2 - Metzler, Scott D.

PB - SPIE

ER -