Can AI provide value in molecular imaging?

Can AI applications in molecular imaging prove capable of directly impacting patient care? Does such technology even have a place in nuclear medicine?

Gerold Porenta, MD, PhD, of Ambulatorium Döbling’s Department of Nuclear Medicine in Vienna asked these questions in a recent commentary published in the Journal of Nuclear Medicine, suggesting the technology currently does not have the scientific backing necessary for widespread clinical use.

“Currently there is little scientific evidence that AI applications in molecular imaging and nuclear medicine can assist in improving the quality of life for patients, reduce morbidity or eliminate premature mortality,” Porenta wrote. “However, while the potential value of AI applications appears evident and even cogent, scientific validation is still the missing link that at present impedes large scale clinical adoption of AI imaging applications.”

Large data sets are a critical component to building AI applications—a problem for nuclear medicine, according to Porenta. And ensuring the characteristics, validity and transparency of such data is central, but has yet to be demonstrated, he added.

“In molecular imaging and nuclear medicine large datasets are difficult to collect,” Porenta wrote. “The individual data in most health care systems belong to the patient and thus can only be shared when proper patient consent has been obtained.”

Because algorithms have potential to directly impact care, they need thorough scientific validation similar to drugs and medical devices.

For example, Porenta noted, deep learning that analyzes myocardial perfusion images and recommends for or against coronary revascularization without a physician may need continuous recertification—yet another task difficult for humans and yet to be proven, he wrote.

“The value of AI applications in medical care can only be confirmed when professional guidelines provide recommendations for their use in specific clinical settings and patient populations,” he wrote.

In a final note, Porenta wrote he was ready to “concede” value to AI imaging technology if it participated in the Society of Nuclear Medicine and Molecular Imaging (SNMMI’s) annual imaging challenge and beat at least one human expert. He doesn’t expect this will happen, but hopes it will, Porenta added.

“In summary, currently there is little scientific evidence for value of AI applications in medical imaging,” he noted. “Several AI applications have received FDA approval but this does not imply that AI applications at present are relevant for medical practice, or that their value has been firmly established.”