Comparison of the accuracy of human readers versus machine-learning algorithms for pigmented skin lesion classification: an open, web-based, international, diagnostic study - 12/06/19

Doi : 10.1016/S1470-2045(19)30333-X

Philipp Tschandl, PhD ^a, Noel Codella, PhD ^c, Bengü Nisa Akay, MD ^d, Giuseppe Argenziano, ProfPhD ^e, Ralph P Braun, MD ^f, Horacio Cabo, ProfMD ^g, David Gutman, PhD ^h, Allan Halpern, ProfMD ⁱ, Brian Helba, BS ^j, Rainer Hofmann-Wellenhof, MD ^k, Aimilios Lallas, PhD ^l, Jan Lapins, MD ^m, Caterina Longo, PhD ^n,^o, Josep Malvehy, MD ^p,^q, Michael A Marchetti, MD ⁱ, Ashfaq Marghoob, MD ^r, Scott Menzies, ProfMD ^s, Amanda Oakley, MBChB ^t, John Paoli, MD ^u, Susana Puig, MD ^p,^q, Christoph Rinner, PhD ^b, Cliff Rosendahl, PhD ^v, Alon Scope, MD ^x, Christoph Sinz, MD ^a, H Peter Soyer, ProfMD ^w, Luc Thomas, ProfPhD ^y,^z,^aa, Iris Zalaudek, MD ^ab, Harald Kittler, MD ^a,^⁎

^a ViDIR Group, Department of Dermatology, Medical University of Vienna, Vienna, Austria

^b Center for Medical Statistics, Informatics and Intelligent Systems (CeMSIIS), Medical University of Vienna, Vienna, Austria

^c IBM Research AI, T J Watson Research Center, Yorktown Heights, NY, USA

^d Department of Dermatology, Medicine Faculty, Ankara University, Ankara, Turkey

^e Dermatology Unit, University of Campania, Naples, Italy

^f Skin Cancer Center, Department of Dermatology, University Hospital Zürich, Zürich, Switzerland

^g Department of Dermatology, Instituto de Investigaciones Médicas, Buenos Aires, Argentina

^h Department of Neurology, Emory University School of Medicine, Atlanta, GA, USA

ⁱ Dermatology Service, Department of Medicine, Memorial Sloan Kettering Cancer Center, New York, NY, USA

^j Kitware, Clifton Park, NY, USA

^k Department of Dermatology, Medical University Graz, Graz, Austria

^l First Department of Dermatology, Aristotle University, Thessaloniki, Greece

^m Department of Dermatology, Karolinska University Hospital and Karolinska Institutet, Stockholm, Sweden

ⁿ Department of Dermatology, University of Modena and Reggio Emilia, Modena, Italy

^o Azienda Unità Sanitaria Locale—IRCCS di Reggio Emilia, Centro Oncologico ad Alta Tecnologia Diagnostica-Dermatologia, Reggio Emilia, Italy

^p Melanoma Unit, Dermatology Department, Hospital Clínic Barcelona, Universitat de Barcelona, IDIBAPS, Barcelona, Spain

^q Centro de Investigación Biomédica en Red de Enfermedades Rarasd (CIBER ER), Instituto de Salud Carlos III, Barcelona, Spain

^r Memorial Sloan Kettering Cancer Center, Hauppauge, NY, USA

^s Sydney Melanoma Diagnostic Centre & Sydney Medical School, Faculty of Medicine and Health, The University of Sydney, Sydney, NSW, Australia

^t Department of Dermatology, Waikato District Health Board and Waikato Clinical Campus, University of Auckland, Hamilton, New Zealand

^u Department of Dermatology and Venereology, Institute of Clinical Sciences, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden

^v School of Clinical Medicine, University of Queensland, University of Queensland, Brisbane, QLD, Australia

^w Dermatology Research Centre, The University of Queensland Diamantina Institute, University of Queensland, Brisbane, QLD, Australia

^x Medical Screening Institute, Sheba Medical Center and Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel

^y Department of Dermatology, Hospitalier Lyon Sud, Lyon, France

^z Lyon Cancer Research Center INSERM U1052—CNRS UMR5286, Lyon, France

^aa Lyon 1 University, Lyon, France

^ab Dermatology Clinic, Maggiore Hospital, University of Trieste, Trieste, Italy

^* Correspondence to: Dr Harald Kittler, Department of Dermatology, Medical University of Vienna, 1090 Vienna, Austria Department of Dermatology Medical University of Vienna Vienna 1090 Austria

En prensa. Pruebas corregidas por el autor. Disponible en línea desde el Wednesday 12 June 2019

Summary

Background

Whether machine-learning algorithms can diagnose all pigmented skin lesions as accurately as human experts is unclear. The aim of this study was to compare the diagnostic accuracy of state-of-the-art machine-learning algorithms with human readers for all clinically relevant types of benign and malignant pigmented skin lesions.

Methods

For this open, web-based, international, diagnostic study, human readers were asked to diagnose dermatoscopic images selected randomly in 30-image batches from a test set of 1511 images. The diagnoses from human readers were compared with those of 139 algorithms created by 77 machine-learning labs, who participated in the International Skin Imaging Collaboration 2018 challenge and received a training set of 10 015 images in advance. The ground truth of each lesion fell into one of seven predefined disease categories: intraepithelial carcinoma including actinic keratoses and Bowen’s disease; basal cell carcinoma; benign keratinocytic lesions including solar lentigo, seborrheic keratosis and lichen planus-like keratosis; dermatofibroma; melanoma; melanocytic nevus; and vascular lesions. The two main outcomes were the differences in the number of correct specific diagnoses per batch between all human readers and the top three algorithms, and between human experts and the top three algorithms.

Findings

Between Aug 4, 2018, and Sept 30, 2018, 511 human readers from 63 countries had at least one attempt in the reader study. 283 (55·4%) of 511 human readers were board-certified dermatologists, 118 (23·1%) were dermatology residents, and 83 (16·2%) were general practitioners. When comparing all human readers with all machine-learning algorithms, the algorithms achieved a mean of 2·01 (95% CI 1·97 to 2·04; p<0·0001) more correct diagnoses (17·91 [SD 3·42] vs 19·92 [4·27]). 27 human experts with more than 10 years of experience achieved a mean of 18·78 (SD 3·15) correct answers, compared with 25·43 (1·95) correct answers for the top three machine algorithms (mean difference 6·65, 95% CI 6·06–7·25; p<0·0001). The difference between human experts and the top three algorithms was significantly lower for images in the test set that were collected from sources not included in the training set (human underperformance of 11·4%, 95% CI 9·9–12·9 vs 3·6%, 0·8–6·3; p<0·0001).

Interpretation

State-of-the-art machine-learning classifiers outperformed human experts in the diagnosis of pigmented skin lesions and should have a more important role in clinical practice. However, a possible limitation of these algorithms is their decreased performance for out-of-distribution images, which should be addressed in future research.