Cover Image

Effects of label noise on deep learning-based skin cancer classification

Recent studies have shown that deep learning is capable of classifying dermatoscopic images at least as well as dermatologists. However, many studies in skin cancer classification utilize non-biopsy-verified training images. This imperfect ground truth introduces a systematic error, but the effects...

Full description

Saved in:
Bibliographic Details
Main Authors: Hekler, Achim (Author) , Kather, Jakob Nikolas (Author) , Krieghoff-Henning, Eva (Author) , Utikal, Jochen (Author) , Meier, Friedegund (Author) , Gellrich, Frank Friedrich (Author) , Upmeier zu Belzen, Julius (Author) , French, Lars E. (Author) , Schlager, Justin Gabriel (Author) , Ghoreschi, Kamran (Author) , Wilhelm, Tabea (Author) , Kutzner, Heinz (Author) , Berking, Carola (Author) , Heppt, Markus V. (Author) , Haferkamp, Sebastian (Author) , Sondermann, Wiebke (Author) , Schadendorf, Dirk (Author) , Schilling, Bastian (Author) , Izar, Benjamin (Author) , Maron, Roman C. (Author) , Schmitt, Max (Author) , Fröhling, Stefan (Author) , Lipka, Daniel (Author) , Brinker, Titus Josef (Author)
Format: Article (Journal)
Language:English
Published: 06 May 2020
In: Frontiers in medicine
Year: 2020, Volume: 7, Pages: 1-7
ISSN:2296-858X
DOI:10.3389/fmed.2020.00177
Online Access:Resolving-System, kostenfrei: http://dx.doi.org/10.3389/fmed.2020.00177
Verlag, lizenzpflichtig, Volltext: https://doi.org/10.3389/fmed.2020.00177
Verlag, lizenzpflichtig, Volltext: https://www.frontiersin.org/articles/10.3389/fmed.2020.00177/full
Get full text
Author Notes:Achim Hekler, Jakob N. Kather, Eva Krieghoff-Henning, Jochen S. Utikal, Friedegund Meier, Frank F. Gellrich, Julius Upmeier zu Belzen, Lars French, Justin G. Schlager, Kamran Ghoreschi, Tabea Wilhelm, Heinz Kutzner, Carola Berking, Markus V. Heppt, Sebastian Haferkamp, Wiebke Sondermann, Dirk Schadendorf, Bastian Schilling, Benjamin Izar, Roman Maron, Max Schmitt, Stefan Fröhling, Daniel B. Lipka and Titus J. Brinker
Description
Summary:Recent studies have shown that deep learning is capable of classifying dermatoscopic images at least as well as dermatologists. However, many studies in skin cancer classification utilize non-biopsy-verified training images. This imperfect ground truth introduces a systematic error, but the effects on classifier performance are currently unknown. Here, we systematically examine the effects of label noise by training and evaluating convolutional neural networks (CNN) with 804 images of melanoma and nevi labeled either by dermatologists or by biopsy. The CNNs are evaluated on a test set of 384 images by means of 4-fold cross validation comparing the outputs with either the corresponding dermatological or the biopsy-verified diagnosis. With identical ground truths of training and test labels, high accuracies with 75.03% (95% CI: 74.39-75.66%) for dermatological and 73.80% (95% CI: 73.10-74.51%) for biopsy-verified labels can be achieved. However, if the CNN is trained and tested with different ground truths, accuracy drops significantly to 64.53% (95% CI: 63.12-65.94%, p<0.01) on a non-biopsy-verified and to 64.24% (95% CI: 62.66-65.83%, p<0.01) on a biopsy-verified test set. In conclusion, deep learning methods for skin cancer classification are highly sensitive to label noise and future work should use biopsy-verified training images to mitigate this problem.
Physical Description:Online Resource
ISSN:2296-858X
DOI:10.3389/fmed.2020.00177

Similar Items