AI-driven cardiac PET attenuation correction shows promise

An AI-driven, cardiac PET Rb-82 PET myocardial perfusion imaging (MPI) exam attenuation correction (AC) algorithm shows comparable performance and diagnostic accuracy to CT-based AC.

The research results were shared by Aakash Dhananjay Shanbhag, PhD, of Cedars Sinai Medical Center in New York City, on June 2 at the 2026 Society of Nuclear Medicine and Molecular Imaging (SNMMI) annual meeting in Los Angeles.

AC is essential for accurate quantification in PET MPI, Shanbhag explained, noting that traditionally the exam is performed with the aid of additional CT scan -- which can introduce CT-PET misregistration artifacts. But to make things trickier, a significant number of PET MPI studies are still performed on stand-alone systems that use “costly and low-quality line source–based transmission scanning for AC,” he and colleagues wrote.

The group developed an AI-based framework using a conditional generative adversarial network for PET AC which they named DeepAC. They trained the model on Rb-82 PET images (with and without AC, n = 1,104) and comparable CT exams, then validated it on an independent dataset from a separate site (n = 1,766). Finally, the group assessed agreement between DeepAC and CT-based AC using summed stress scores (SSS) and type of exam’s diagnostic performance for identifying obstructive coronary artery disease (CAD) -- defined as equal to or more than 50% left main coronary artery stenosis or 70% stenosis in other heart vessels -- using total perfusion deficit (TPD).

The team found the following:

• DeepAC showed closer agreement to clinical summed stress scores compared to actual AC (p < 0.001).
• Correlation of quantitative DeepAC and AC summed stress scores was excellent in 11 of 17 AHA segments (all mid- and apical regions), with good to excellent correlation in basal segments (p < 0.001 for all).
• In cases with invasive angiography (n = 278), area under the curve (AUC) was similar between CT-based AC and DeepAC total perfusion deficit (0.77 and 0.76 respectively; p = 0.45).

Cardiac stress test imaging comparison showing DeepAC and CT-based attenuation correction scans for two patients with quantitative perfusion data.Aakash Dhananjay Shanbhag, PhD, and SNMMI

The takeaway? DeepAC could be a “practical alternative in cardiac PET,” Shanbhag concluded.

Check out AuntMinnie’s full coverage of SNMMI 2026 on our ShowCast.