Multicenter evaluation of interpretable AI for coronary artery disease diagnosis from PET biomarkers - Scorecard - MDSpire

Multicenter evaluation of interpretable AI for coronary artery disease diagnosis from PET biomarkers

  • By

  • Wenhao Zhang

  • Jacek Kwiecinski

  • Aakash Shanbhag

  • Robert J. H. Miller

  • Shiva Mostafavi

  • Giselle Ramirez

  • Jirong Yi

  • Donghee Han

  • Damini Dey

  • Dominika Grodecka

  • Kajetan Grodecki

  • Mark Lemley

  • Paul Kavanagh

  • Joanna X. Liang

  • Jianhang Zhou

  • Valerie Builoff

  • Jon Hainer

  • Sylvain Carre

  • Leanne Barrett

  • Andrew J. Einstein

  • Stacey Knight

  • Steve Mason

  • Viet T. Le

  • Wanda Acampa

  • Samuel Wopperer

  • Panithaya Chareonthaitawee

  • Daniel S. Berman

  • Marcelo F. Di Carli

  • Piotr J. Slomka

  • January 14, 2026

  • 0 min

Share

Clinical Scorecard: Assessment of Interpretable Artificial Intelligence for Diagnosing Coronary Artery Disease Using PET Biomarkers Across Multiple Centers

At a Glance

CategoryDetail
ConditionCoronary Artery Disease (CAD)
Key MechanismsIntegration of PET myocardial perfusion imaging (MPI) parameters including perfusion imaging, myocardial blood flow (MBF), myocardial flow reserve (MFR), coronary artery calcium (CAC), and functional metrics using AI
Target PopulationPatients undergoing PET/CT myocardial perfusion imaging for suspected CAD
Care SettingMulticenter clinical imaging settings with PET/CT MPI capability

Key Highlights

  • AI model integrates 10 common PET MPI parameters to improve CAD diagnosis beyond individual measurements.
  • Model validated externally on 1278 patients from three institutions, demonstrating robust generalizability.
  • AI model outperformed clinical scores and individual PET biomarkers (ischemic TPD, MFR, stress MBF, CAC) with AUC of 0.83 vs. 0.80 or lower.

Guideline-Based Recommendations

Diagnosis

  • Use PET/CT MPI combining perfusion imaging, MBF, MFR, and CAC for comprehensive CAD assessment.
  • Consider AI-based integration of PET MPI parameters to improve diagnostic accuracy and risk stratification.

Management

  • Leverage automated CAC scoring from CT attenuation correction scans routinely acquired during PET MPI.
  • Incorporate AI model outputs to identify high-risk patients for targeted clinical decision-making.

Monitoring & Follow-up

  • Monitor PET MPI parameters including ischemic TPD, stress MBF, MFR, and CAC for changes over time.
  • Use AI model thresholds to track disease progression or response to therapy.

Risks

  • Recognize that individual PET parameters alone may underperform compared to integrated AI analysis.
  • Ensure quality control of PET/CT imaging and data completeness to maintain AI model accuracy.

Patient & Prescribing Data

1664 patients with suspected CAD; median age 68 years, 65% male

AI model identifies more high-risk patients than conventional quantitative thresholds, supporting improved clinical risk stratification and potential treatment tailoring.

Clinical Best Practices

  • Utilize tracer-specific normal limits and dedicated kinetic models for perfusion and flow quantification to ensure data compatibility.
  • Impute missing PET MPI data conservatively using median values or related scores to maintain dataset integrity.
  • Apply AI models based on standard PET imaging measurements rather than raw images to enhance interpretability and clinical adoption.
  • Validate AI models externally across multiple centers and tracers to confirm robustness and generalizability.

References

Original Source(s)

Multicenter evaluation of interpretable AI for coronary artery disease diagnosis from PET biomarkers

  • by Wenhao Zhang, Jacek Kwiecinski, Aakash Shanbhag, Robert J. H. Miller, Shiva Mostafavi, Giselle Ramirez, Jirong Yi, Donghee Han, Damini Dey, Dominika Grodecka, Kajetan Grodecki, Mark Lemley, Paul Kavanagh, Joanna X. Liang, Jianhang Zhou, Valerie Builoff, Jon Hainer, Sylvain Carre, Leanne Barrett, Andrew J. Einstein, Stacey Knight, Steve Mason, Viet T. Le, Wanda Acampa, Samuel Wopperer, Panithaya Chareonthaitawee, Daniel S. Berman, Marcelo F. Di Carli, Piotr J. Slomka

  • January 14, 2026

Related Content

AI Scribes Lag Clinicians on Note Quality

A VHA study across 11 vendors finds AI-generated primary care notes score lower than clinician-written notes, with the largest deficits in thoroughness, organization, and usefulness

Retinal Age Model Tied to Disease Risk

Fundus-based tool estimates biologic aging but shows modest gains over prior models

Application and prospect of artificial intelligence in diagnostic imaging of prostate cancer