AI techniques in medical imaging reconstruction could lead to incorrect diagnoses: study

Study data published in PNAS suggest that machine learning and artificial intelligence (AI) are highly unstable in medical image reconstruction and may lead to false positive and false negative results. A team of researchers designed a series of tests for medical image reconstruction algorithms based on AI and deep learning, and found that these methods result in myriad artefacts, among other major errors in the final images, whereas the effects were not typically present in non-AI-based imaging.

Study co-lead Anders Hansen said "we've found that AI techniques are highly unstable in medical imaging, so that small changes in the input may result in big changes in the output." He explained that "there is a limit to how good a reconstruction can be given restricted scan time. In some sense, modern AI techniques break this barrier, and as a result become unstable."  

For the study, investigators designed a stability test with algorithms and software to detect flaws AI-based medical imaging systems, including magnetic resonance imaging, computerised tomography and nuclear magnetic resonance. The researchers focused on instabilities associated with tiny perturbations, instabilities with respect to small structural changes, such as a brain image with or without a small tumour, and instabilities tied to changes in the number of samples.

Results showed that certain tiny movements led to many artefacts in the final images, details were blurred or completely removed, and that image reconstruction quality deteriorated with repeated subsampling. "These errors were widespread across the different types of neural networks," the authors noted.

"When it comes to critical decisions around human health, we can't afford to have algorithms making mistakes," said Hansen, adding that "we found that the tiniest corruption, such as may be caused by a patient moving, can give a very different result if you're using AI and deep learning to reconstruct medical images, meaning that these algorithms lack the stability they need."

The researchers are looking to provide the fundamental limits to what can be done with AI techniques in medical imaging reconstruction, in order to understand which problems can be solved. 

Did you like this article?