Skip to main content

November 2019 - Philipp Tschandl

Priv.Doz. Dr. med.univ. Philipp Tschandl, PhD

MedUni Wien RESEARCHER OF THE MONTH November 2019

Expert-Level Diagnosis of Nonpigmented Skin Cancer by Combined Convolutional Neural Networks

Almost any cancer of the skin can present as a non-pigmented lesion, the most common being basal cell carcinoma and sub-forms of squamous cell carcinoma, which are highly prevalent in an elderly caucasian population. While dermatoscopy enhances the diagnostic accuracy of clinicians in distinguishing benign from malignant skin tumors, evaluation of non-pigmented lesions remains difficult, especially for non-experts [6]. Thus, an automated decision support for this type of lesion is desirable, to potentially enable non-experts to better manage non-pigmented skin lesions themselves.
Convolutional neural networks (CNN) are one type of machine learning algorithms with high accuracy in the classification of digital images. In this work CNNs were trained for classifying either a clinical or dermatoscopic image and merged to a final prediction model via gradient boosting. This combined CNN (cCNN) model was compared to 95 human raters, including more than 60 dermatologists, for the detection of nonpigmented skin cancer in more than 2000 unknown cases. The cCNN achieved a diagnostic accuracy superior to beginners (<3 years of experience in dermatoscopy) and intermediates (3-10 years), and not significantly different to expert (>10 years) raters. Overall, data indicate a promising future for CNN application in skin cancer detection, but limitations in generalizability and scope of integrated disease classes demand further work before reaching clinical applicability.

Selected Literature

  1. Tschandl P, Rosendahl C, Akay BN, Argenziano G, Blum A, Braun RP, Cabo H, Gourhant JY, Kreusch J, Lallas A, Lapins J, Marghoob A, Menzies S, Neuber NM, Paoli J, Rabinovitz HS, Rinner C, Scope A, Soyer HP, Sinz C, Thomas L, Zalaudek I, Kittler H. Expert-Level Diagnosis of Nonpigmented Skin Cancer by Combined Convolutional Neural Networks. JAMA Dermatology 2019;155(1):58-65.
  2. Tschandl P, Sinz C, Kittler H. Domain-specific classification-pretrained fully convolutional network encoders for skin lesion segmentation. Comput Biol Med 2019;104:111-116.
  3. Tschandl P, Argenziano G, Razmara M, Yap J. Diagnostic Accuracy of Content Based Dermatoscopic Image Retrieval with Deep Classification Features. Br J Dermatol 2018
  4. Tschandl P, Rosendahl C, Kittler H. The HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions. Sci Data 2018
  5. Weinbacher A, Tschandl P, Kittler H, Rinner C. Interoperable Localisation of Lesions on the Human Skin. Stud Health Technol Inform. 2018;247:850-854.
  6. Sinz C, Tschandl P, Rosendahl C, Akay BN, Argenziano G, Blum A, Braun RP, Cabo H, Gourhant JY, Kreusch J, Lallas A, Lapins J, Marghoob AA, Menzies SW, Paoli J, Rabinovitz HS, Rinner C, Scope A, Soyer HP, Thomas L, Zalaudek I, Kittler H.Accuracy of dermatoscopy for the diagnosis of nonpigmented cancers of the skin. J Am Acad Dermatol. 2017;77(6):1100-1109.
  7. Tschandl P, Kittler H, Argenziano G. A pretrained neural network shows similar diagnostic accuracy to medical students in categorizing dermatoscopic images after comparable training conditions. Br J Dermatol. 2017;177(3):867-869.
  8. Tschandl P, Hofmann L, Fink C, Kittler H, Haenssle HA. Melanomas vs. nevi in high-risk patients under long-term monitoring with digital dermatoscopy: do melanomas and nevi already differ at baseline? J Eur Acad Dermatol Venereol. 2017;31(6):972-977.

Priv.Doz. Dr. Philipp Tschandl

Medizinische Universität Wien
Universitätsklinik für Dermatologie
Währinger Gürtel 18-20
1090 Wien

T: +43 (0)1 40400-77000
philipp.tschandl@meduniwien.ac.at
https://www.meduniwien.ac.at/vidir