RT Journal Article SR Electronic T1 Automatic Lung Lobe Analysis with COVID-19 Support JF Journal of Nuclear Medicine JO J Nucl Med FD Society of Nuclear Medicine SP 146 OP 146 VO 62 IS supplement 1 A1 Fei Gao A1 Zhoubing Xu A1 Sasa Grbic A1 Bruce Spottiswoode YR 2021 UL http://jnm.snmjournals.org/content/62/supplement_1/146.abstract AB 146Objectives: Coronavirus disease 2019 (COVID-19) is an infectious lung disease affecting more than 90 million individuals worldwide as of Jan 11, 2021, and still significantly impacts our daily life. In 2020, we presented an automatic lung segmentation workflow to analyze lung function quantitatively at the lung lobe level [1]. To further support and contribute to COVID-19 analysis, we extended our work in [1] and developed a fully automatic research prototype analysis workflow using the deep learning based segmentation [2] trained for lung fissure delineation and opacity detection with over 8000 chest CT image volumes. The objective of this study is to evaluate the performance of the proposed workflow with publicly available datasets. Methods: The RSNA International COVID-19 Open Radiology Database (RICORD) provided imaging data with annotations and supporting clinical information for education and research [3]. In this study, we used the first batch of 120 datasets from RICORD to evaluate our workflow. One dataset is of irregular size and was excluded from the analysis. The remaining datasets comprised 43 datasets scanned with General Electric (GE) Healthcare scanners, 18 datasets scanned with Siemens Healthineers scanners, and 58 datasets scanned with scanners from undisclosed vendors. For annotation, the reconstructed slice thickness was 3 mm for 76 datasets, 2 mm for 33 datasets, and between 2 and 3 mm for the remaining 10 datasets. The datasets were split to 3 parts and annotated by 3 teams respectively. In our evaluation, we segmented both lung lobes and infectious opacity for each dataset, then compared the infectious opacity with the annotation. Following the RICORD annotation Quality Assurance (QA) check, 104 of the 119 datasets with adequate image quality were included for quantitative and statistical analysis using two types of correlations, Pearson’s coefficient, and Kendall’s Tau. The result from datasets with inadequate quality was also provided for comparison. We also evaluated the inter-team variability of annotation. Results: The automatic lung analysis workflow completed successfully for all patient datasets. For the 104 datasets with adequate image quality, the mean and standard deviation of the percentage of infectious opacity in the whole lung (segmented in our workflow) is 31.8±26% for our segmentation and 33.5±30% for the annotation. The Pearson’s coefficient is 0.765 (p<.0001) and the Kendall’s Tau is 0.720 (p<.0001). The results are then grouped by the annotation team. The Pearson’s coefficients are 0.765(p<.0001) for team 1, 0.670(p<.0001) for team 2, and 0.827(p<.0001) for team 3. The Kendall’s Tau is 0.704(p<.0001) for team 1, 0.608(p<.0001) for team 2, and 0.778(p<.0001) for team 3. The segmentation results from our analysis workflow have the highest correlation with the annotation from team 3, while the lowest correlation with the annotation from team 1. For the datasets with inadequate quality noted in RICORD QA check (breathing motion, artifacts, etc.), the Pearson’s coefficient is decreased to 0.429(p=0.188) and the Kendall’s Tau is decreased to 0.348(p=0.206). Conclusions: The image quality plays a critical role in the accuracy of segmentation. With these completely unseen public datasets from multiple vendors and relatively suboptimal reconstruction quality, our automatic segmentation shows reasonable and acceptable results compared with the RICORD manual annotation. The inter-team variability of annotation is observed and will also impact clinical reading. This evaluation demonstrates that our extended automatic lung analysis workflow has the ability to provide robust automatic lung lobe segmentation with COVID-19 analysis support for SPECT/CT and PET/CT lung ventilation and perfusion studies. [1] Gao, Fei, et al. JNM 61.supplement 1 (2020): 1489-1489. [2] Chaganti, Shikha, et al. Radiology: Artificial Intelligence 2.4 (2020): e200048. [3] Tsai EB, et al. Radiology 2021 (in press).