Stress testing reveals gaps in clinic readiness of image-based diagnostic artificial intelligence models
Open Access
- 21 January 2021
- journal article
- research article
- Published by Springer Science and Business Media LLC in npj Digital Medicine
- Vol. 4 (1), 1-8
- https://doi.org/10.1038/s41746-020-00380-6
Abstract
Artificial intelligence models match or exceed dermatologists in melanoma image classification. Less is known about their robustness against real-world variations, and clinicians may incorrectly assume that a model with an acceptable area under the receiver operating characteristic curve or related performance metric is ready for clinical use. Here, we systematically assessed the performance of dermatologist-level convolutional neural networks (CNNs) on real-world non-curated images by applying computational “stress tests”. Our goal was to create a proxy environment in which to comprehensively test the generalizability of off-the-shelf CNNs developed without training or evaluation protocols specific to individual clinics. We found inconsistent predictions on images captured repeatedly in the same setting or subjected to simple transformations (e.g., rotation). Such transformations resulted in false positive or negative predictions for 6.5–22% of skin lesions across test datasets. Our findings indicate that models meeting conventionally reported metrics need further validation with computational stress tests to assess clinic readiness.Keywords
Funding Information
- Melanoma Research Alliance (622732, 622732)
- UCSF Helen Diller Family Comprehensive Cancer Center
- UCSF Summer Explore Fellowship, Marguerite Schoeneman Award, Alameda-Contra Costa Medical Association Summer Fellowship, UCSF/UCB Joint Medical Program Thesis Grant
- Doris Duke Charitable Foundation
This publication has 32 references indexed in Scilit:
- Automated Dermatological Diagnosis: Hype or Reality?Journal of Investigative Dermatology, 2018
- A Novel Contributory Cross-Domain Group Password-Based Authenticated Key Exchange Protocol with Adaptive SecurityPublished by Institute of Electrical and Electronics Engineers (IEEE) ,2017
- Dermatologist-level classification of skin cancer with deep neural networksNature, 2017
- Deep Residual Learning for Image RecognitionPublished by Institute of Electrical and Electronics Engineers (IEEE) ,2016
- ImageNet Large Scale Visual Recognition ChallengeInternational Journal of Computer Vision, 2015
- PH2 - A dermoscopic image database for research and benchmarking2018 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 2013
- Assessing the Performance of Prediction ModelsEpidemiology, 2010
- Research electronic data capture (REDCap)—A metadata-driven methodology and workflow process for providing translational research informatics supportJournal of Biomedical Informatics, 2008
- Comparing the Areas under Two or More Correlated Receiver Operating Characteristic Curves: A Nonparametric ApproachBiometrics, 1988