Use of neural networks for improved diagnosis of pulmonary artery embolism using ventilation/perfusion SPECT/CT and the possible omission of the ventilation component

Pulmonary artery embolism is a common and potentially life-threatening disease whose diagnosis and course assessment are of great importance. The clinical presentation is highly variable, ranging from asymptomatic patients to spontaneous death. To prevent a serious course and complications, a quick diagnosis and a quick start of therapy are essential. Despite optimal care, late complications such as recurrent disease and chronic thromboembolic pulmonary hypertension are possible. Typical algorithms recommend ventilation/perfusion SPECT (V/P-SPECT) and computed tomographic angiography of the pulmonary arteries (CTPA) as imaging diagnostics. The advantages of V/P-SPECT are that it can be carried out more frequently due to fewer contraindications, higher sensitivity with comparable specificity, lower radiation exposure and the possibility of quantifying the affected lung parenchyma, which is important for follow-up examinations. However, the availability of the study is lower. The ventilation SPECT component in particular requires expensive radionuclide generators, which limits the range of investigations. In addition, the ventilation SPECT is also critical in relation to the ongoing COVID-19 pandemic, since its implementation exposes medical staff to a high risk of infection when examining infected patients. An alternative is to perform a pure perfusion SPECT without a ventilation SPECT. In this way, perfusion failures can be detected, and the ventilated lung parenchyma can be assessed using a CT, which is acquired using hybrid imaging technology in the same examination (SPECT/CT). In this way, a comparably high sensitivity can be achieved as for V/P-SPECT, but the specificity is significantly reduced, which can lead to a high rate of false-positive findings and an overdiagnosis of pulmonary artery embolism. There is therefore still an unmet clinical need for an improvement in V/P-SPECT and the pure perfusion SPECT/CT examination. The high number of recent publications shows growing scientific interest in artificial intelligence (AI)-based methods for the analysis of nuclear medicine images by neural networks, to enable automatic image evaluation and improved diagnostics. So far there is only insufficient data about such applications on V/P-SPECT data for the diagnosis of pulmonary artery embolism. The hypothesis of this project is therefore that a highly developed evaluation of V/P-SPECT/CT using AI-based methods has the potential to significantly increase the diagnostic quality and also the specificity of pure perfusion SPECT/CT for diagnosis of a pulmonary artery embolism to such an extent that a ventilation SPECT becomes unnecessary.