Automatic extraction of left-atrial volume from a clinical perfusion 15O-water PET

Objectives: nlargement of the left atrium is associated with increased left-ventricular (LV) filling pressures and poor outcome, regardless of factors such as LV ejection fraction or coronary flow reserve. LA volume (LAV) measurements are recommended by guidelines for several cardiac diseases. The aim of this study was to develop and validate an automated method of extracting left-atrial volumes from a standard cardiac perfusion PET scan.

Methods: An algorithm was created for automatically segmenting the left atrium using the 4D information present in any dynamic PET scan, in this study using 15O-water as perfusion tracer. For validation, PET-LAV was compared to echo-derived LAV in 36 patients with systolic heart failure (HF) and to MRI-derived LAV in 14 patients from a cohort of 35 patients with moderate-severe mitral regurgitation (MR). As PET-LAV represents the mean LAV during the cardiac cycle, for echo, mean LAV was calculated as (1/3)[asterisk](2[asterisk]LAV_diastolic+LAV_systolic) whilst for MRI, only LAV_systolic was available. Test-retest repeatability was assessed in 15 patients referred for evaluation of ischemic heart disease (IHD), scanned twice during the same day. The method was then applied to the remaining 21 patients with MR, 14 patients with cardiac amyloidosis (CA) and, additionally, 121 consecutive patients referred for clinical assessment of myocardial blood flow. Clinical patients were grouped based on their hyperemic MBF response: normals (no documented IHD at follow up, MBF (in mL∙min^-1∙g^-1) > 2.3 in all segments and global CFR > 2.5), 1V (single vessel disease), 3V (multi-vessel disease, global MBF < 2.3 or CFR between 1.5-2.5 and MVD (severe microvascular dysfunction, globally reduced CFR < 1.5). Results: PET-derived LAV correlated highly with echo-derived LAV (r2=0.77, p<0.001, mean difference 20±16 mL) and with MRI LAV (r2=0.68, p<0.001, mean difference -66±33 mL), there was however with a systematic difference compared to both echocardiography and MRI (p<0.001 for both). Test-retest repeatability of the method was high (intraclass correlation coefficient = 0.92). For normals, LAV corrected for body-surface area (LAVI) was 26.7±7.0 ml∙m^-2, in line with reference values for echocardiography and MRI. LAVI was increased for 3V (31.3±9.2, p=0.02), MVD (36.5±14.9, p=0.02), HF (42.8±18.0, p<0.001), MI (58.8±21.8, p<0.001) and CA (36.1±11.8, p=0.01) but not for 1V (28.0±9.6, p=0.54). LAVI in clinical patients correlated weakly with stress MBF (r2=0.09, p<0.001) and CFR (r2=0.05, p=0.003).

Conclusions: Left-atrial volumes can be obtained fully automatically from dynamic 15O-water PET with high accuracy and reproducibility. This method does not require any additional scans and can routinely be added to standard MBF protocols. Finally, increased LAV is associated with cardiomyopathy rather than coronary artery disease.