Do We Still Need Human Input?

An MRI scan read in 4 seconds? Are we hearing this right?

You are, and welcome to the present and the future of automated machine learning programs that have the ability to significantly increase the speed of analysis of specialized MRI scans.

Don’t worry though—it’s not ready for prime time just yet!

Now, new research sheds light on just far we have come in terms of development of such machine learning programs.

According to a new study published in the journal, Circulation: Cardiovascular Imaging, analysis of cardiac MRI scans using automated machine learning can be performed significantly faster and with comparable accuracy to human interpretation by trained cardiologists.

It generally takes about 13 minutes for a trained physician (cardiologist) to interpret a cardiac MRI. But with the advent of artificial intelligence (AI) and machine learning algorithms, a scan can be analyzed with similar accuracy in approximately four seconds–nearly 186 times faster!

Cardiac MRI scans help cardiologists make preoperative decisions for surgical planning by providing important measurements of heart structure and function.

But they may also help to make decisions related to timing of cardiac surgery, implantation of defibrillators and continuing or discontinuing infusions of cardiotoxic chemotherapy for cancer patients. Such clinical decisions ultimately rely on accurate and precise measurements from cardiac MRI scans.  

Acceleration of data gleaned from such MRI scans might also help to improve speed of clinical decision-making and subsequent health outcomes, ultimately affecting care from a population health perspective.

The study was performed in the UK, where approximately 150,000 cardiac MRI scans are performed annually. Extrapolating the number of scans performed per year, the investigators calculated that using AI to interpret scans could potentially result in saving 54 clinician-days annually at each health facility.

For their study, researchers developed and trained a neural network to read the cardiac MRI scans of nearly 600 patients. They found that there was no significant difference in accuracy when the machine learning algorithm was tested for precision compared to an expert and trainee on 110 separate patients from multiple centers.

“Cardiovascular MRI offers unparalleled image quality for assessing heart structure and function; however, current manual analysis remains basic and outdated. Automated machine learning techniques offer the potential to change this and radically improve efficiency, and we look forward to further research that could validate its superiority to human analysis,” said study author Charlotte Manisty, M.D. Ph.D. in a press release. “Our dataset of patients with a range of heart diseases who received scans enabled us to demonstrate that the greatest sources of measurement error arise from human factors. This indicates that automated techniques are at least as good as humans, with the potential soon to be ‘super-human’—transforming clinical and research measurement precision.”

Another expert believes that such novel technology offers promise for both clinical and research-based applications.

“Machine learning will impact how we read cardiac MRI studies in our day-to day work and how we train future generations of physicians,” said Jennifer Conroy, M.D., Associate Director of Cardiac CT and MRI, Lenox Hill Hospital, Northwell Health. “Adoption of novel tools that increase precision and efficiency are useful not only for individual patient care but also in performing larger scale research studies.”

What’s clear from the study is that this particular machine learning algorithm did not perform better compared with human interpretation. While it was significantly faster, the question arises whether human input is still necessary.

I firmly believe that we are not at a point where machines can go unchecked and unsupervised after they deliver an interpretation. Clinicians still need to provide oversight and verify readings to assure patient safety and quality of care. While reduction in clinical hours is certainly an added benefit to such technology, we must first assure that there is no patient harm in larger scale studies.

That said, the study does highlight the potential utility of AI as a transformative technique in increasing speed without sacrificing accuracy and precision for interpretation of complex imaging scans for patients with structural heart disease.  

More studies and investigations will certainly be required before this type of approach could become a viable option for healthcare systems.

Bottom line: we won’t be replacing cardiologists with AI and machine learning algorithms anytime soon.