Purpose: Cancer patients are vulnerable to COVID-19 and other infections. Daily CBCT is common for image-guided radiotherapy. To protect patients and staff lung infections must be detected timely, which is possible on daily thoracic CBCT. However, it is time consuming to inspect all slices of a thoracic CBCT, especially for longitudinal changes. Here we develop and optimise a tool for rapid detection lung of infections by visual comparison of daily images projected in 2D.
Methods: All daily thoracic CBCTs are processed overnight by a pipeline containing registration, cropping to lungs, filtration, and maximum intensity projection. Parameters were optimised visually using 150 cases. A visual timeline of the treatment is presented in a report and animation. Images are reviewed each morning by a radiographer who flags issues to a clinical oncologist to determine if further action is required. The tool was retrospectively evaluated in 285 patients treated between January and June 2020 and prospectively implemented in a clinical workflow and departmental quality system in July 2020.
Results: Rigid registration outperformed deformable registration, the latter being affected by lung lesions. Optimal cropping was the planning CT lungs - 3mm, balancing sensitivity for peripheral lesions and rejection of pleural effusion changes. Blurring (σ=1cm) in AP direction suppressed healthy lung structures while maintaining infection contrast. Report generation took ~60s per case, and inspection ~12s. To date, over 400 patients were prospectively evaluated and 4 asymptomatic patients were diagnosed with COVID-19 based on CBCT imaging changes. A further patient was identified as having a non COVID-19 lung infection on CBCT.
Conclusion: We developed and validated a near real-time tool to identify lung density changes on CBCT indicative of COVID-19 or other lung infections. CBCT imaging provides a unique opportunity to study temporal progression of lung infections and protect patients and staff.