Artificial Intelligence to Optimize Pulmonary Embolism Diagnosis During Covid-19 Pandemic by Perfusion SPECT/CT, a Pilot Study

INTRODUCTION: COVID-19 pneumonia is associated with a high rate of pulmonary embolism (PE) and its diagnosis can be changeling. In patients with contraindications for CT pulmonary angiography (CTPA) or nondiagnostic on CTPA, perfusion single photon emission computed tomography/computed tomography (Q-SPECT/CT) is a diagnosis option in patients with COVID-19 disease. Our aim was to develop an artificial intelligence (AI) model based on QSPECT/ CT images of patients during the COVID-19 pandemic which is able to classify lung lesions to optimize PE diagnosis. METHODS: Single center study with a prospective observational branch with patients who tested positive for COVID-19 and underwent a perfusion SPECT-CT study for diagnosis of PE, from April to September 2020. The second branch is retrospective, with patients in pre-pandemic period who underwent Q-SPECT/CT for diagnosis of PE from January to December 2018. For each patient, a Q-SPECT/CT based on intravenous administration of 6 mCi (222MBq) of 99mTC-macroaggregates of human-albumin (99mTC-MAA) was performed, with the subsequent tomoscintigraphy (SPECT) and a CT. The entire image pre-processing (volume exploration, segmentation analysis, registration analysis) was conducted with MATLAB. The final diagnosis for each patient and the different tissue type segments were analyzed and validated by two senior nuclear physicians. Our Intelligent Radiomic System for the identification of patients with PE (with or without pneumonia) is based on a local analysis of SPECT-CT volumes that considers both CT and SPECT values for each volume point. Volumes are first co-registered and their intensity is normalized in [0, 1] to account for differences due to acquisition. A support vector machine (SVM) model was trained to discriminate among 4 different types of tissue: no pneumonia without PE (control group), no pneumonia with PE, pneumonia without PE and pneumonia with PE. We followed a k-fold (with k=30) scheme for statistical analysis of results. RESULTS: We collected 133 patients, 63 in prospective branch (26 with PE, 22 without PE, 15 indeterminate- PE) and 70 in retrospective branch (31 healthy/control, 39 PE). Concerning the local analysis, for all cases we obtain an average sensitivity and positive predictive value over 92%. CONCLUSION: This study represents a first step towards a complete intelligent radiomic system to optimize the diagnosis of PE by Q-SPECT/CT. The capability to detect alterations in perfusion for COVID-19 pneumonia encourages developing a tool in the cloud for clinical use and further investigates if it can also predict long term complications.