Enhanced classifier training to improve precision of a convolutional neural network to identify images of skin lesions

• Published in: PloS one 2019-06, Vol.14 (6), p.e0218713-e0218713
• Main Authors: Brinker, Titus J, Hekler, Achim, Enk, Alexander H, von Kalle, Christof
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial neural networks; Biology and Life Sciences; Cancer; Classification; Classifiers; Computer and Information Sciences; Conferences and conventions; Databases, Factual; Deep Learning; Dermatologists; Dermatology; Dermoscopy; Diagnosis; Disclosure of information; Humans; Image classification; Image Interpretation, Computer-Assisted - methods; Lesions; Medical diagnosis; Medical imaging; Medical research; Medicine and Health Sciences; Melanoma; Melanoma - classification; Melanoma - diagnostic imaging; Neural networks; Neural Networks, Computer; Patient outcomes; Pattern recognition; Reproducibility of Results; Research and Analysis Methods; Skin; Skin - diagnostic imaging; Skin cancer; Skin diseases; Skin Neoplasms - classification; Skin Neoplasms - diagnostic imaging; Social Sciences; Systematic review; Test sets; Training
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0218713

In recent months, multiple publications have demonstrated the use of convolutional neural networks (CNN) to classify images of skin cancer as precisely as dermatologists. However, these CNNs failed to outperform the International Symposium on Biomedical Imaging (ISBI) 2016 challenge which ranked the average precision for classification of dermoscopic melanoma images. Accordingly, the technical progress represented by these studies is limited. In addition, the available reports are impossible to reproduce, due to incomplete descriptions of training procedures and the use of proprietary image databases or non-disclosure of used images. These factors prevent the comparison of various CNN classifiers in equal terms. To demonstrate the training of an image-classifier CNN that outperforms the winner of the ISBI 2016 CNNs challenge by using open source images exclusively. A detailed description of the training procedure is reported while the used images and test sets are disclosed fully, to insure the reproducibility of our work. Our CNN classifier outperforms all recent attempts to classify the original ISBI 2016 challenge test data (full set of 379 test images), with an average precision of 0.709 (vs. 0.637 of the ISBI winner) and with an area under the receiver operating curve of 0.85. This work illustrates the potential for improving skin cancer classification with enhanced training procedures for CNNs, while avoiding the use of costly equipment or proprietary image data.