Although numerous recent publications and preprints report machine learning models with high performance at this task 3, 4, 5, 6, 7, 8, the trustworthiness of these models needs to be evaluated rigorously before deployment in a clinical setting 9. The prospect of applying artificial neural networks to the detection of COVID-19 in chest radiographs has generated interest from machine learning (ML) researchers and radiologists alike, given its potential to (1) help guide management in resource-limited settings that lack sufficient numbers of the gold-standard polymerase chain reaction with reverse transcription (RT-PCR) assay and (2) clarify cases of suspected false negatives from the RT-PCR assay 1, 2. These findings demonstrate that explainable AI should be seen as a prerequisite to clinical deployment of machine-learning healthcare models. In addition, we show that evaluation of a model on external data is insufficient to ensure AI systems rely on medically relevant pathology, because the undesired ‘shortcuts’ learned by AI systems may not impair performance in new hospitals. Because this approach to data collection has also been used to obtain training data for the detection of COVID-19 in computed tomography scans and for medical imaging tasks related to other diseases, our study reveals a far-reaching problem in medical-imaging AI. We observe that the approach to obtain training data for these AI systems introduces a nearly ideal scenario for AI to learn these spurious ‘shortcuts’. Using state-of-the-art techniques in explainable AI, we demonstrate that recent deep learning systems to detect COVID-19 from chest radiographs rely on confounding factors rather than medical pathology, creating an alarming situation in which the systems appear accurate, but fail when tested in new hospitals. However, the robustness of these systems remains unclear. Artificial intelligence (AI) researchers and radiologists have recently reported AI systems that accurately detect COVID-19 in chest radiographs.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |