Systematic outperformance of 112 dermatologists in multiclass skin cancer image classification by convolutional neural networks

• Verfasst von: Maron, Roman C. [VerfasserIn]
• Verfasst von: Weichenthal, Michael [VerfasserIn]
• Verfasst von: Utikal, Jochen [VerfasserIn]
• Verfasst von: Hekler, Achim [VerfasserIn]
• Verfasst von: Berking, Carola [VerfasserIn]
• Verfasst von: Hauschild, Axel [VerfasserIn]
• Verfasst von: Enk, Alexander [VerfasserIn]
• Verfasst von: Haferkamp, Sebastian [VerfasserIn]
• Verfasst von: Klode, Joachim [VerfasserIn]
• Verfasst von: Schadendorf, Dirk [VerfasserIn]
• Verfasst von: Jansen, Philipp [VerfasserIn]
• Verfasst von: Holland-Letz, Tim [VerfasserIn]
• Verfasst von: Schilling, Bastian [VerfasserIn]
• Verfasst von: Kalle, Christof von [VerfasserIn]
• Verfasst von: Fröhling, Stefan [VerfasserIn]
• Verfasst von: Gaiser, Maria [VerfasserIn]
• Verfasst von: Hartmann, Daniela [VerfasserIn]
• Verfasst von: Gesierich, Anja Heike [VerfasserIn]
• Verfasst von: Kähler, Katharina C. [VerfasserIn]
• Verfasst von: Wehkamp, Ulrike [VerfasserIn]
• Verfasst von: Karoglan, Ante [VerfasserIn]
• Verfasst von: Bär, Claudia [VerfasserIn]
• Verfasst von: Brinker, Titus Josef [VerfasserIn]
• Titel: Systematic outperformance of 112 dermatologists in multiclass skin cancer image classification by convolutional neural networks
• Verf.angabe: Roman C. Maron, Michael Weichenthal, Jochen S. Utikal, Achim Hekler, Carola Berking, Axel Hauschild, Alexander H. Enk, Sebastian Haferkamp, Joachim Klode, Dirk Schadendorf, Philipp Jansen, Tim Holland-Letz, Bastian Schilling, Christof von Kalle, Stefan Fröhling, Maria R. Gaiser, Daniela Hartmann, Anja Gesierich, Katharina C. Kähler, Ulrike Wehkamp, Ante Karoglan, Claudia Bär, Titus J. Brinker, Collabrators
• E-Jahr: 2019
• Jahr: 14 August 2019
• Umfang: 9 S.
• Fussnoten: Gesehen am 06.11.2019
• Titel Quelle: Enthalten in: European journal of cancer
• Ort Quelle: Amsterdam [u.a.] : Elsevier, 1992
• Jahr Quelle: 2019
• Band/Heft Quelle: 119(2019), Seite 57-65
• ISSN Quelle: 1879-0852
• DOI: doi:10.1016/j.ejca.2019.06.013
• Datenträger: Online-Ressource
• Sprache: eng
• Sach-SW: Artificial intelligence
• Sach-SW: Melanoma
• Sach-SW: Skin cancer
• Sach-SW: Skin cancer screening
• K10plus-PPN: 168116910X

Abstract

Background - Recently, convolutional neural networks (CNNs) systematically outperformed dermatologists in distinguishing dermoscopic melanoma and nevi images. However, such a binary classification does not reflect the clinical reality of skin cancer screenings in which multiple diagnoses need to be taken into account. - Methods - Using 11,444 dermoscopic images, which covered dermatologic diagnoses comprising the majority of commonly pigmented skin lesions commonly faced in skin cancer screenings, a CNN was trained through novel deep learning techniques. A test set of 300 biopsy-verified images was used to compare the classifier's performance with that of 112 dermatologists from 13 German university hospitals. The primary end-point was the correct classification of the different lesions into benign and malignant. The secondary end-point was the correct classification of the images into one of the five diagnostic categories. - Findings - Sensitivity and specificity of dermatologists for the primary end-point were 74.4% (95% confidence interval [CI]: 67.0-81.8%) and 59.8% (95% CI: 49.8-69.8%), respectively. At equal sensitivity, the algorithm achieved a specificity of 91.3% (95% CI: 85.5-97.1%). For the secondary end-point, the mean sensitivity and specificity of the dermatologists were at 56.5% (95% CI: 42.8-70.2%) and 89.2% (95% CI: 85.0-93.3%), respectively. At equal sensitivity, the algorithm achieved a specificity of 98.8%. Two-sided McNemar tests revealed significance for the primary end-point (p < 0.001). For the secondary end-point, outperformance (p < 0.001) was achieved except for basal cell carcinoma (on-par performance). - Interpretation - Our findings show that automated classification of dermoscopic melanoma and nevi images is extendable to a multiclass classification problem, thus better reflecting clinical differential diagnoses, while still outperforming dermatologists at a significant level (p < 0.001).