Deep learning models for electrocardiograms are susceptible to adversarial attack
| Autor: | Lior Jankelson, Larry A. Chinitz, Xintian Han, Rajesh Ranganath, Luca Foschini, Yuxuan Hu |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
0301 basic medicine
Arrhythmia detection Computer science Machine learning computer.software_genre Article General Biochemistry Genetics and Molecular Biology Electrocardiography 03 medical and health sciences Adversarial system Deep Learning 0302 clinical medicine Humans Artificial neural network business.industry Deep learning General Medicine Models Theoretical ComputingMethodologies_PATTERNRECOGNITION 030104 developmental biology 030220 oncology & carcinogenesis Deep neural networks Neural Networks Computer Artificial intelligence Ecg signal business Raw data computer Classifier (UML) Algorithms |
| Zdroj: | Nat Med |
| ISSN: | 1546-170X 1078-8956 |
| DOI: | 10.1038/s41591-020-0791-x |
| Popis: | Electrocardiogram (ECG) acquisition is increasingly widespread in medical and commercial devices, necessitating the development of automated interpretation strategies. Recently, deep neural networks have been used to automatically analyze ECG tracings and outperform physicians in detecting certain rhythm irregularities1. However, deep learning classifiers are susceptible to adversarial examples, which are created from raw data to fool the classifier such that it assigns the example to the wrong class, but which are undetectable to the human eye2,3. Adversarial examples have also been created for medical-related tasks4,5. However, traditional attack methods to create adversarial examples do not extend directly to ECG signals, as such methods introduce square-wave artefacts that are not physiologically plausible. Here we develop a method to construct smoothed adversarial examples for ECG tracings that are invisible to human expert evaluation and show that a deep learning model for arrhythmia detection from single-lead ECG6 is vulnerable to this type of attack. Moreover, we provide a general technique for collating and perturbing known adversarial examples to create multiple new ones. The susceptibility of deep learning ECG algorithms to adversarial misclassification implies that care should be taken when evaluating these models on ECGs that may have been altered, particularly when incentives for causing misclassification exist. The development of an algorithm that can imperceptibly manipulate electrocardiographic data to fool a deep learning model for diagnosing cardiac arrhythmia highlights the potential vulnerability of artificial intelligence-enabled diagnosis to adversarial attacks. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full text from SpringerLink