The dynamics and prognostic value of FDG PET-metrics in weekly monitoring of (chemo)radiotherapy for NSCLC

Early radiologic and metabolic tumour response assessment during combined chemo-radiotherapy for locally advanced NSCLC

Background

The role of early treatment response for patients with locally advanced non-small cell lung cancer (LA-NSCLC) treated with concurrent chemo-radiotherapy (cCRT) is unclear. The study aims to investigate the predictive value of response to induction chemotherapy (iCX) and the correlation with pattern of failure (PoF).

Materials and methods

Patients with LA-NSCLC treated with cCRT were included for analyses (n = 276). Target delineations were registered from radiotherapy planning PET/CT to diagnostic PET/CT, in between which patients received iCX. Volume, sphericity, and SUVpeak were extracted from each scan. First site of failure was categorised as loco-regional (LR), distant (DM), or simultaneous LR+M (LR+M). Fine and Gray models for PoF were performed: a baseline model (including performance status (PS), stage, and histology), an image model for squamous cell carcinoma (SCC), and an image model for non-SCC. Parameters included PS, volume (VOL) of tumour, VOL of lymph nodes, ΔVOL, sphericity, SUVpeak, ΔSUVpeak, and oligometastatic disease.

Results

Median follow-up was 7.6 years. SCC had higher sub-distribution hazard ratio (sHR) for LRF (sHR = 2.771 [1.577:4.87], p < 0.01) and decreased sHR for DM (sHR = 0.247 [0.125:0.485], p  <  0.01). For both image models, high diagnostic SUVpeak increased risk of LRF (sHR = 1.059 [1.05:1.106], p < 0.01 for SCC, sHR = 1.12 [1.03:1.21], p < 0.01 for non-SCC). Patients with SCC and less decrease in VOL had higher sHR for DM (sHR = 1.025[1.001:1.048] pr. % increase, p = 0.038).

Conclusion

Poor response in disease volume was correlated with higher sHR of DM for SCC, no other clear correlation of response and PoF was observed. Histology significantly correlated with PoF with SCC prone to LRF and non-SCC prone to DM as first site of failure. High SUVpeak at diagnosis increased the risk of LRF for both histologies.

Short 2-[18F]Fluoro-2-Deoxy-D-Glucose PET Dynamic Acquisition Protocol to Evaluate the Influx Rate Constant by Regional Patlak Graphical Analysis in Patients With Non-Small-Cell Lung Cancer

Purpose: To test a short 2-[¹⁸F]Fluoro-2-deoxy-D-glucose (2-[¹⁸F]FDG) PET dynamic acquisition protocol to calculate K_i using regional Patlak graphical analysis in patients with non-small-cell lung cancer (NSCLC).

Methods: 24 patients with NSCLC who underwent standard dynamic 2-[¹⁸F]FDG acquisitions (60 min) were randomly divided into two groups. In group 1 (n = 10), a population-based image-derived input function (pIDIF) was built using a monoexponential trend (10–60 min), and a leave-one-out cross-validation (LOOCV) method was performed to validate the pIDIF model. In group 2 (n = 14), K_i was obtained by standard regional Patlak plot analysis using IDIF (0–60 min) and tissue response (10–60 min) curves from the volume of interests (VOIs) placed on descending thoracic aorta and tumor tissue, respectively. Moreover, with our method, the Patlak analysis was performed to obtain K_i,s using IDIF_Fitted curve obtained from PET counts (0–10 min) followed by monoexponential coefficients of pIDIF (10–60 min) and tissue response curve obtained from PET counts at 10 min and between 40 and 60 min, simulating two short dynamic acquisitions. Both IDIF and IDIF_Fitted curves were modeled to assume the value of 2-[¹⁸F]FDG plasma activity measured in the venous blood sampling performed at 45 min in each patient. Spearman's rank correlation, coefficient of determination, and Passing–Bablok regression were used for the comparison between K_i and K_i,s. Finally, K_i,s was obtained with our method in a separate group of patients (group 3, n = 8) that perform two short dynamic acquisitions.

Results: Population-based image-derived input function (10–60 min) was modeled with a monoexponential curve with the following fitted parameters obtained in group 1: a = 9.684, b = 16.410, and c = 0.068 min⁻¹. The LOOCV error was 0.4%. In patients of group 2, the mean values of K_i and K_i,s were 0.0442 ± 0.0302 and 0.33 ± 0.0298, respectively (R² = 0.9970). The Passing–Bablok regression for comparison between K_i and K_i,s showed a slope of 0.992 (95% CI: 0.94–1.06) and intercept value of −0.0003 (95% CI: −0.0033–0.0011).

Conclusions: Despite several practical limitations, like the need to position the patient twice and to perform two CT scans, our method contemplates two short 2-[¹⁸F]FDG dynamic acquisitions, a population-based input function model, and a late venous blood sample to obtain robust and personalized input function and tissue response curves and to provide reliable regional K_i estimation.