Collective human intelligence outperforms artificial intelligence in a skin lesion classification task

Background and objectives Convolutional neural networks (CNN) enable accurate diagnosis of medical images perform on or above the level individual physicians. Recently, collective human intelligence (CoHI) was shown to exceed diagnostic accuracy individuals. Thus, performance CoHI (120 dermatologists) versus dermatologists two state-of-the-art CNN investigated. Patients Methods Cross-sectional reader study with presentation 30 clinical cases 120 dermatologists. Six diagnoses were offered votes collected via remote voting devices (quizzbox®, Quizzbox Solutions GmbH, Stuttgart, Germany). Dermatoscopic classified by a binary multiclass (FotoFinder Systems Bad Birnbach, Three sets classifications scored against ground truth: (1) CoHI, (2) dermatologists, (3) CNN. Results attained significantly higher [95 % confidence interval] (80.0 [62.7 %–90.5 %]) than (75.7 [73.8 %–77.5 (70.0 [52.1 %–83.3 %]; all P < 0.001) in classifications. Moreover, achieved sensitivity (82.4 [59.0 %–93.8 specificity (76.9 [49.7 %–91.8 (sensitivity 77.8 [75.3 %–80.2 %], 73.0 [70.6 %–75.4 70.6 [46.9 %–86.7 69.2 [42.4 %–87.3 %]). The superior that (P evaluation, latter comparable Conclusions Our analysis revealed majority vote an interconnected group outperformed individuals demanding skin lesion classification task.