DOI: 10.1161/circ.148.suppl_1.18984 ISSN: 0009-7322

Abstract 18984: Artificial Intelligence-Based Automated Left Ventricle Mass Quantification From Non-Contrast Cardiac CT Scans: Correlation With Contrast CT and Cardiac MRI

Donghee Han, Aakash Shanbhag, Aditya Killekar, Parker B Waechter, Nicholas Kwok, Daniel Berman, Piotr J Slomka
  • Physiology (medical)
  • Cardiology and Cardiovascular Medicine

Background: Non-contrast CT scans are currently not used to assess left ventricular myocardial mass (LV mass), which is usually evaluated with contrast CT or cardiac magnetic resonance (CMR). We aimed to assess the feasibility of LV mass estimation from non-contrast CT calcium scans using an artificial intelligence (AI) approach and compare it to measurements from contrast CT and CMR.

Methods: A total of 316 patients (Age: 57.1±16.7, 53.2% male) who underwent coronary CT angiography (with non-contrast calcium CT and contrast CT) and CMR were included. The median interval between CT and CMR was 22 days (IQR: 3-76). An nn-Unet AI model (Total Segmentator), previously trained on data from independent sites, automatically segmented non-contrast CT structures including the myocardium. LV mass was also measured from contrast CT and CMR by experts during routine clinical reporting.

Results: The AI segmentation took ~22 seconds per case. There was no significant difference in LV mass between the AI and contrast CT (136.5±55.3 vs. 139.6±58.0 g, p=0.114). An excellent correlation was observed between AI and contrast CT (r=0.84, p<0.001) (Figure). Bland-Altman analysis showed a minimal bias of 3.15. Compared to CMR, both AI and non-contrast CT showed higher LV mass (136±55.3, 139.6±58.0 vs. 127.1±53.1 g, respectively, both p<0.001). The correlations of AI and contrast CT with CMR were excellent with no significant differences. (r= 0.86 and 0.85, both p<0.001). Bland-Altman analysis indicated small bias toward higher LV mass with AI and contrast CT (-9.35 and -12.5, respectively) compared to CMR.

Conclusions: The AI-based automated LV mass from non-contrast CT showed excellent correlations and minimal biases when compared to LV mass measured from contrast CT and CMR. This fully automated measure could be applied for routine LV mass evaluation in patients with a non-contrast CT scan, providing an additional biomarker previously thought not measurable by this modality.

More from our Archive