FDG-PET/CT for lung cancer: phantom quality-control for respiratory motion compensation protocols, AI-algorithm for tumor segmentation and delta-radiomic model for recurrence prediction

Ziel/Aim The evaluation of lung tumors with PET/CT imaging presents challenges due to respiratory movement. Respiratory motion compensation (4D) imaging protocols could lead to an improvement on the image quality. The aim of this study is to provide open-source tools to optimize the implementation of FDG-PET in lung cancer and to facilitate the evaluation of its impact in clinical practice.

Methodik/Methods We developed an open-source package for the quality-control of motion compensated protocols, based on experimental dynamic (QUASAR-platform) phantom imaging. PET image quality was assessed in terms of resolution (Jazyscakc-phantom), accuracy in volume (NEMA-phantom) and accuracy in activity concentration (Density-phantom with fillable inserts), by the recovery coefficients (RC) with respect to the ideal response of imaging the phantom without movement. We additionally developed an AI tumor segmentation algorithm based on nn-Unet, trained with 560 PET images (4D-PETs for 56 patients) and validated with 80 images (8 patients). We developed a delta-radiomic model for recurrence prediction with 47 lung cancer patients, by employing radiomic variability across 4D-PETs as a patient individualized normalization factor to emphasize statistically relevant changes during treatment.

Ergebnisse/Results The quality-control package provided RC values of 102 PET images in less than 5 min and allowed the identification of the best protocol for the TF-64 Philips PET/CT, with (RCresol=0.6, RCvolume=1.6, RCactivity=0.90). The Dice-Similarity-Coefficient=0.71 for AI-segmentation was better than the inter-observer variability. Radiomic-model predicted recurrence with an Area-Unther-the-Curve of 0.80 and 0.63 for training and validation cohort, respectively.

Schlussfolgerungen/Conclusions We developed different open-source tools to facilitate and to optimize the implementation of FDG-PET in lung cancer clinical practice.

Publication History

Article published online:
30 March 2023