RT Journal Article
SR Electronic
T1 Artificial Intelligence based segmental quantification of pulmonary perfusion for pre-transplant workup.
JF Journal of Nuclear Medicine
JO J Nucl Med
FD Society of Nuclear Medicine
SP 1698
OP 1698
VO 62
IS supplement 1
A1 Kritika Subramanian
A1 Sujoy Dhar
A1 Lady Sawoszczyk
A1 Fei Gao
A1 Adam Leep
A1 Ramya Rajaram
A1 John Hornung
A1 Joseph Osborne
A1 Elisabeth O'Dwyer
YR 2021
UL http://jnm.snmjournals.org/content/62/supplement_1/1698.abstract
AB 1698Objectives: Lung ventilation/perfusion scintigraphy is routinely performed to evaluate for lung function prior to transplant along with computed tomography (CT) of the chest and spirometry to aid surgical planning1. Lung scintigraphy in the context of pre-operative evaluation is beneficial as it provides an estimation of lung function, evaluates cardiac shunts, and can detect chronic thromboembolic pulmonary hypertension, this identifying the more diseased lung2. This is especially important in the setting of bilateral lung transplant as it allows the surgeon to determine which lung should be transplanted first3 in single lung and lobar transplantation. Lung transplantation is traditionally performed in children and adults with severe end-stage lung disease (e.g. Chronic Obstructive Pulmonary Disease (COPD) and pulmonary fibrosis) and are increasing in the context of improved life expectancy and increasing organ donor scarcity4,5. In response to this Siemens has developed a fully automatic research prototype algorithm for lung analysis called LungVQ6 that uses artificial intelligence to assist with accurately identifying pulmonary lobes for pre-operative perfusion imaging. This study describes the first clinical experience of this software. Methods: Three patients who were being evaluated for lung transplant underwent perfusion SPECT (Single Photon Emission Computed Tomography)/CT as per departmental protocol with Tc99m- macro-aggregated albumin (MAA) (4mCi/148MBq intravenous) on a GE Optima scanner. “Pseudoplanar” images were created from SPECT data, with quantification of perfusion in anterior and posterior projection of both lungs, using the geometric mean 2D analysis (automatic GE software using standard 6-lobe algorithm that splits each lung in to three sections), providing radiotracer activity in counts and percentage. Concurrently, the novel Siemens LungVQ algorithm was applied on the same data to perform a 3-D analysis using a 5-lobe segmentation method and the results were compared to the standard 2-D analysis. Results: Both standard geometric mean analysis and the Siemens LungVQ algorithm demonstrated equivalent total percentage of MAA radiotracer activity in both lungs (p=0.314) (Figure 1). However Siemens LungVQ algorithm additionally provided lobar volumes and accurately identified pulmonary fissures and thus lobar boundaries as validated by CT findings. This was reiterated by the significant difference in perfusion findings of the right upper (p=0.039) and right middle (p=0.0004) lobes, which did correspond to the change in calculated lobar size/volume. To note, the CT data provided to the Siemens LungVQ algorithm was 2.5mm thick slices while recommended slice size is less than 2 mm. The recommended CT kernel is B40 while the data from the GE Optima scanner used a smoother kernel. Study interpretation was also limited by breathing artifacts which may have resulted in segmentation artifacts. Conclusion: The MAA percentage activity values calculated from perfusion SPECT/CT using the standard GE method and the novel Siemens LungVQ algorithm were similar. The Siemens LungVQ algorithm also correctly identified pulmonary lobes, which may assist in pre- transplant workup especially in lobar lung transplantation. Pending further validation in a larger cohort of patients, the Siemens LungVQ algorithm could guide better identification of patients for transplantation and the most appropriate surgical approach.