Advanced search
1 file | 469.05 KB Add to list

Quantifying uncertainty of deep neural networks in skin lesion classification

Pieter Van Molle (UGent) , Tim Verbelen (UGent) , Cedric De Boom, Bert Vankeirsbilck (UGent) , Jonas De Vylder (UGent) , Bart Diricx, Tom Kimpe (UGent) , Pieter Simoens (UGent) and Bart Dhoedt (UGent)
Author
Organization
Abstract
Deep neural networks are becoming the new standard for automated image classification and segmentation. Recently, such models are also gaining traction in the context of medical diagnosis. However, when using a neural network as a decision support tool, it is important to also quantify the (un)certainty regarding the outputs of the system. Current Bayesian techniques approximate the true predictive output distribution via sampling, and quantify the uncertainty based on the variance of the output samples. In this paper, we highlight the limitations of a variance based metric, and propose a novel uncertainty metric based on the overlap of the output distributions. We show that this yields promising results on the HAM10000 dataset for skin lesion classification.
Keywords
Deep learning, Uncertainty, Dermatology, Skin lesions

Downloads

  • (...).pdf
    • full text (Published version)
    • |
    • UGent only
    • |
    • PDF
    • |
    • 469.05 KB

Citation

Please use this url to cite or link to this publication:

MLA
Van Molle, Pieter, et al. “Quantifying Uncertainty of Deep Neural Networks in Skin Lesion Classification.” UNCERTAINTY FOR SAFE UTILIZATION OF MACHINE LEARNING IN MEDICAL IMAGING AND CLINICAL IMAGE-BASED PROCEDURES, edited by H. Greenspan et al., vol. 11840, Springer, 2019, pp. 52–61, doi:10.1007/978-3-030-32689-0_6.
APA
Van Molle, P., Verbelen, T., De Boom, C., Vankeirsbilck, B., De Vylder, J., Diricx, B., … Dhoedt, B. (2019). Quantifying uncertainty of deep neural networks in skin lesion classification. In H. Greenspan, R. Tanno, & M. Erdt (Eds.), UNCERTAINTY FOR SAFE UTILIZATION OF MACHINE LEARNING IN MEDICAL IMAGING AND CLINICAL IMAGE-BASED PROCEDURES (Vol. 11840, pp. 52–61). https://doi.org/10.1007/978-3-030-32689-0_6
Chicago author-date
Van Molle, Pieter, Tim Verbelen, Cedric De Boom, Bert Vankeirsbilck, Jonas De Vylder, Bart Diricx, Tom Kimpe, Pieter Simoens, and Bart Dhoedt. 2019. “Quantifying Uncertainty of Deep Neural Networks in Skin Lesion Classification.” In UNCERTAINTY FOR SAFE UTILIZATION OF MACHINE LEARNING IN MEDICAL IMAGING AND CLINICAL IMAGE-BASED PROCEDURES, edited by H. Greenspan, R. Tanno, and M. Erdt, 11840:52–61. Springer. https://doi.org/10.1007/978-3-030-32689-0_6.
Chicago author-date (all authors)
Van Molle, Pieter, Tim Verbelen, Cedric De Boom, Bert Vankeirsbilck, Jonas De Vylder, Bart Diricx, Tom Kimpe, Pieter Simoens, and Bart Dhoedt. 2019. “Quantifying Uncertainty of Deep Neural Networks in Skin Lesion Classification.” In UNCERTAINTY FOR SAFE UTILIZATION OF MACHINE LEARNING IN MEDICAL IMAGING AND CLINICAL IMAGE-BASED PROCEDURES, ed by. H. Greenspan, R. Tanno, and M. Erdt, 11840:52–61. Springer. doi:10.1007/978-3-030-32689-0_6.
Vancouver
1.
Van Molle P, Verbelen T, De Boom C, Vankeirsbilck B, De Vylder J, Diricx B, et al. Quantifying uncertainty of deep neural networks in skin lesion classification. In: Greenspan H, Tanno R, Erdt M, editors. UNCERTAINTY FOR SAFE UTILIZATION OF MACHINE LEARNING IN MEDICAL IMAGING AND CLINICAL IMAGE-BASED PROCEDURES. Springer; 2019. p. 52–61.
IEEE
[1]
P. Van Molle et al., “Quantifying uncertainty of deep neural networks in skin lesion classification,” in UNCERTAINTY FOR SAFE UTILIZATION OF MACHINE LEARNING IN MEDICAL IMAGING AND CLINICAL IMAGE-BASED PROCEDURES, Shenzhen, China, 2019, vol. 11840, pp. 52–61.
@inproceedings{8637791,
  abstract     = {{Deep neural networks are becoming the new standard for automated image classification and segmentation. Recently, such models are also gaining traction in the context of medical diagnosis. However, when using a neural network as a decision support tool, it is important to also quantify the (un)certainty regarding the outputs of the system. Current Bayesian techniques approximate the true predictive output distribution via sampling, and quantify the uncertainty based on the variance of the output samples. In this paper, we highlight the limitations of a variance based metric, and propose a novel uncertainty metric based on the overlap of the output distributions. We show that this yields promising results on the HAM10000 dataset for skin lesion classification.}},
  author       = {{Van Molle, Pieter and Verbelen, Tim and De Boom, Cedric and Vankeirsbilck, Bert and De Vylder, Jonas and Diricx, Bart and Kimpe, Tom and Simoens, Pieter and Dhoedt, Bart}},
  booktitle    = {{UNCERTAINTY FOR SAFE UTILIZATION OF MACHINE LEARNING IN MEDICAL IMAGING AND CLINICAL IMAGE-BASED PROCEDURES}},
  editor       = {{Greenspan, H. and Tanno, R. and Erdt, M.}},
  isbn         = {{9783030326883}},
  issn         = {{0302-9743}},
  keywords     = {{Deep learning,Uncertainty,Dermatology,Skin lesions}},
  language     = {{eng}},
  location     = {{Shenzhen, China}},
  pages        = {{52--61}},
  publisher    = {{Springer}},
  title        = {{Quantifying uncertainty of deep neural networks in skin lesion classification}},
  url          = {{http://doi.org/10.1007/978-3-030-32689-0_6}},
  volume       = {{11840}},
  year         = {{2019}},
}

Altmetric
View in Altmetric
Web of Science
Times cited: