A pilot study on lung cancer detection based on regional metabolic activity distribution in digital low-dose 18F-FDG PET.

Objectives: To investigate the potential of automatic lung cancer detection on submillisievert dose F-fludeoxyglucose (F-FDG) scans using different positron emission tomography (PET) parameters, as a primary step towards a potential new indication for F-FDG PET in lung cancer screening.

Methods: We performed a retrospective cohort analysis with 83 patients referred for F-FDG PET/CT, including of 34 patients with histology-proven lung cancer and 49 patients without lung disease. Aside clinical standard PET images (PET) two additional low-dose PET reconstructions were generated, using only 15 s and 5 s of the 150 s list mode raw data of the full-dose PET, corresponding to 10% and 3.3% of the original F-FDG activity. The lungs were subdivided into three segments on each side, and each segment was classified as normal or containing cancer. The following standardized uptake values (SUVs) were extracted from PET per lung segment: SUV, SUV, SUV, SUV and SUV. A multivariate linear regression model was used and cross-validated. The accuracy for lung cancer detection was tested with receiver operating characteristics analysis and T-statistics was used to calculate the weight of each parameter.

Results: The T-statistics showed that SUV was the most important discriminative factor for lung cancer detection. The multivariate model achieved an area under the curve of 0.97 for full-dose PET, 0.85 for PET with PET reconstructions resulting in a still high sensitivity the PET reconstruction of 80%.

Conclusion: This pilot study indicates that segment-based, quantitative PET parameters of low-dose PET reconstructions could be used to automatically detect lung cancer with high sensitivity.

Advances In Knowledge: Automated assessment of PET parameters in low-dose PET may aid for an early detection of lung cancer.