E2133. Validation of Artificial Intelligence-Based Measurements for Cardiac Volume in CT Compared to ECG
  1. Christopher Fan; University of Texas Southwestern
  2. Michael Long; University of Texas Southwestern
  3. Suhny Abbara; University of Texas Southwestern
  4. Ron Peshock; University of Texas Southwestern
  5. Fernando Kay; University of Texas Southwestern
Cardiovascular (CV) diseases remain the leading cause of global mortality, necessitating early detection with medical imaging. Medical imaging provides the opportunity to detect CV diseases before they become clinically apparent. Notably, computed tomography (CT) has become a widely utilized method to investigate cardiothoracic pathologies. Many studies have shown that volumetric quantification of the heart with CT is accurate and correlated with chamber dilation or cardiomegaly when compared to gold standards such as MRI and ECG. However, visual assessment of cardiac volume is inconsistent and manual measurement of cardiac CTs is time-consuming. This has led to the recent development of artificial intelligence (AI) tools to assess volume rapidly. While artificial intelligence (AI) has increasingly been employed to automatically extract quantitative biomarkers from CTs, their accuracy and reliability in clinical settings against established methods for assessing cardiac size warrant further exploration. The aim of this study was to test the correlation of cardiac volume measured by an AI algorithm with ground truth evaluation by ECG in clinical patients.

Materials and Methods:
The HIPAA-compliant and IRB-approved study included 130 patients from a single university institution who underwent a chest CT and an echo within 1 month of each other. Total cardiac volume (TCVAI) was automatically measured using AI-Rad Companion (Siemens). Clinical echo reports were reviewed for the presence of dilation of any of the four chambers or left ventricular hypertrophy (LVH). Demographic parameters and history of CV disease were recorded as well as a secondary TCVAI from a chest CT scan closest to the first CT.

The cohort’s mean age was 62.3 ± 12.3 years and 58.5% were men. 85.4% had a prior history of CV disease. TCVAI was larger in patients with CV when compared with patients without CV disease (<em>p</em> = 0.008.). TCVAI was significantly larger in patients with any chamber dilation (range, smallest to largest p among chambers), but no association was found with LVH (<em>p</em> = 0.335). TCVAI was associated with the number of dilated cardiac chambers on echo after adjustments for the presence of cardiovascular disease, sex, age, height, weight, and number of days between CT and echo (<em>p</em> - value < 0.001). The presence of CV disease was associated with the number of dilated chambers on ECG (<em>p</em> - value = 0.023).

Our results support that cardiac volume measurements automatically extracted from routine chest CT using AI can serve as a surrogate for cardiac enlargement and were significantly larger in patients with CV diseases. These results also indicated that the Siemens AI was accurate when accounting for other demographic factors such as sex, age, height, weight, and time. Further studies are needed to establish normal reference ranges in clinical populations, which could be used to identify patients with subclinical CV disease.