The prognostic value of F18 Fluorothymidine positron emission tomography for assessing the response of malignant pleural mesothelioma to chemotherapy – A prospective cohort study

Malignant pleural mesothelioma is difficult to prognosticate. F18‐Fluorodeoxyglucose positron emission tomography (FDG PET) shows promise for response assessment but is confounded by talc pleurodesis. F18‐Fluorothymidine (FLT) PET is an alternative tracer specific for proliferation. We compared the prognostic value of FDG and FLT PET and determined the influence of talc pleurodesis on these parameters.


Introduction
Malignant pleural mesothelioma (MPM) is a cancer with a poor prognosis. 1The 'Modified Response Evaluation Criteria in Solid Tumours' (mRECIST) is the standard for measuring treatment response but has poor specificity. 2,35][6] FDG PET specificity for MPM is reduced by inflammatory tissue, arising secondary to talc pleurodesis and comorbid lung diseases. 7,8Figure 1 illustrates the difficulties in FDG PET interpretation.
FLT is a tracer that has been assessed for diagnosing malignancy and monitoring treatment response.Radiomics can be extracted from FLT with total lesion proliferation (TLP = MTV 9 SUVMean) analogous to FDG TLG.Buck et al. 9 and Yap et al. 10 found in non-small cell lung cancer FLT SUV better correlated with the Ki-67 proliferation index than FDG SUV.Pio et al. 11 compared FDG and FLT in monitoring treatment response in breast cancer and found that changes in FLT uptake better correlated with long-term treatment efficacy than FDG.In fibrosarcoma tumour models for mice, FLT showed statistical significance for measuring early tumour treatment response to cisplatin. 12FLT has been limited by signal-to-noise ratio from bone marrow activity, with an increased rate of false negatives in the detection of malignancy. 13Figure 1 demonstrates the bone marrow uptake seen on FLT PET.
The MiM software platform has semi-automatic region growing tools for extracting tumour radiomics and can isolate and remove the bone marrow.As FLT has shown promise for monitoring early treatment response and specifically with cisplatin, we aimed to assess the utility of FLT in monitoring treatment response in MPM to cisplatin chemotherapy.Using MiM, we aimed to generate MTVs to extract quantitative radiomics from FLT PET and FDG PET and compare them to qualitative variables such as mRECIST after one cycle of chemotherapy, to determine if FLT monitored treatment response better than FDG, and if FLT is also confounded by talc pleurodesis.
The prespecified hypothesis was that changes in FLT PET radiomics would have greater prognostic value than changes in FDG PET radiomics and mRECIST after one cycle of chemotherapy, and would not be confounded by talc pleurodesis.

Patients
Participants with a histologically or cytologically confirmed diagnosis of MPM, over 18 years of age, European Cooperative Oncology Group (ECOG) performance status 0-2, and disease either measurable or non-measurable on CT were prospectively recruited for the study.Those with prior surgery, chemotherapy or radiotherapy for tumour treatment were excluded.Patients who had previously undergone talc pleurodesis were included.Histology of patient disease was reviewed and classified as biphasic, epithelioid or sarcomatoid.Patients were enrolled over the period from 12 May 2007 to 28 July 2009.In this study, 33 patients with an already confirmed diagnosis of MPM were prospectively recruited over this period forming a consecutive series.The census date was 4 years after the first patient commenced their first cycle of chemotherapy.

Clinical assessment and management
At baseline asbestos exposure, smoking history and ECOG score were recorded.Patients received whole body FLT PET-CT and FDG PET-CT scans, and a helical thoracic CT-scan with 5 mm slices within the first 28 days after enrolment into the study and prior to commencing chemotherapy.CT-scan and PET scans were performed within 14 days of each other.Patients exceeding these time constraints were excluded.FDG scans were always performed before FLT scans.
Patients received a 21-day cycle of chemotherapy consisting of intravenous cisplatin 80 mg/m 2 on day 1, and intravenous pemetrexed 500 mg/m 2 on days 1, 8 and 15.Subsequent FLT PET, FDG PET and CT-scans were taken within 4 days and after day 21 of their chemotherapy cycle.These intervals for the scans were selected to try and reduce the acute inflammatory effects from the administration of chemotherapy that might be detected on FDG PET.Patients with measurable disease and for whom a response to chemotherapy was noted on CTscan using mRECIST were eligible for up to six cycles of treatment.Patients with non-measurable disease on CTscan were scheduled for up to four cycles.The number of cycles was reduced for patients as clinically indicated.PET scan results had no influence on patient treatment.Patients were followed up after completion of treatment until death.

CT protocol
Patients had a contrast enhanced helical thoracic CT-scan with 5 mm slices performed.CT images were acquired on a Siemens (CTIInc., Knoxville, USA) Biograph 16 PET-CT using sequential CT (120 kv) acquisition.

CT qualitative analysis
CT-scans were assessed using mRECIST by a radiologist and blinded to the patient's FLT PET and FDG PET scans.

PET protocols
FLT and FDG PET scans were performed with a Siemens (CTIInc.)Biograph 16 PET-CT, using sequential CT (120 kv) and PET (OSEM reconstruction with time-offlight reconstruction and a 6.0 mm FWHN Gaussian point spread function) acquisition.Patients were fasted for 6 h prior to scanning, and 200 MBq/m 2 of FDG was administered intravenously 60 min prior to scanning in a quiet room.Blood glucose level was measured prior to FDG administration and required to be <10 mmol/L.FLT PET protocol required no specific preparation prior to imaging, and 200 MBq/m 2 of FLT was administered intravenously 1 h prior to scanning.PET scans were performed over multiple bed positions, with 4 min per bed position and 50% overlap per field of view.

PET quantitative analysis
Quantitative analysis of PET scans was performed using the MiM software platform (version 6).FDG PET MTVs were generated using MiM's automated 3D regiongrowing edge-finding algorithm.The lower threshold limit for the algorithm was defined by the liver, chosen for the homogeneity of its metabolic activity.A 3 cm sphere was localised to the right lobe and the mean SUV and standard deviation of the sphere were calculated.MTVs were generated by setting the edge-finding algorithms lower threshold at 2 standard deviations above the sphere's mean SUV.
The algorithm delineated MTVs measured in millilitres (ml) on the patient's FDG PET scans, and we further refined the volumes using MiM's PET Edge tool.This was performed on FDG PET scans at baseline and after one cycle of chemotherapy (C1) and reviewed by a nuclear medicine physician blinded to CT and FLT PET results.The MiM Hounsfield unit guided bone mapping tool was used to isolate bone volumes on CT.
Figures 2 and 3 illustrate the novel workflows used to generate FLT PET MTVs.Transposing the Hounsfield guided bone map onto FLT allowed for subtraction of bone marrow uptake from FLT MTVs.The FDG MTVs transposed onto the FLT scans were further refined using MiM's PET Edge tool to remove areas of the MTV that did not correspond with areas of FLT uptake and expanded to include areas of increased FLT uptake.The final FLT MTV was therefore generated by subtracting the Hounsfield mapped bone uptake from the transposed FDG volumes that had been modified using the PET Edge tool.
The following radiomics were extracted from FDG tumour volumes at baseline and C1; tumour SUVmean, tumour SUVmax, MTV and TLG.An absolute change in these variables after one cycle of chemotherapy was defined as (C1 -Baseline), and a percentage change was defined as (C1-Baseline)/(Baseline).SUV values were standardised by patient weight.For FLT the same radiomics were extracted except TLP was extracted instead of TLG.

Statistical analysis
Patient overall survival time was calculated from commencement of chemotherapy (day 1) until date of death.Survival analysis was performed to correlate qualitative and quantitative variables with patient overall survival time.Univariate Cox regression analysis was conducted on all variables to determine hazard ratios (HR), 95% confidence intervals (CI), and P-values.Those with P ≤ 0.05 were considered statistically significant for an exploratory analysis.Any variables with P > 0.05 were considered insignificant.The same criteria were applied to all variables and considered exploratory.Significant variables on univariate analysis were implemented into a multivariate Cox regression analysis to account for correlations and co-dependency between the different variables.Analysis was performed for the entire cohort, the talc pleurodesis subset, and the non-talc pleurodesis subset.Kaplan-Meier curves were generated for statistically significant variables on multivariate analysis.Cut-offs for the Kaplan-Meier curves were selected to split the dataset into three sub-groups.The first sub-group included all values showing an increase in measures of tumour volume or metabolic activity.The final two sub-groups were separated along a value close to the median of the remaining data to make for similarly sized groups.Log-rank testing was performed for the Kaplan-Meier curves.Statistical analysis was performed using IBM SPSS Statistics for Windows, Version 25.0.(Armonk, NY: IBM Corp).Table 1 summarises cohort demographics, with CT Stage from the start of treatment.Median overall survival time was 350 days, and the shortest was 29 days.Tables 2 and 3 summarise the P-values, hazard ratios, and 95% confidence intervals for quantitative FDG and FLT variables, and qualitative variables respectively.

Study population
Univariate FDG survival analysis P-values degraded for the talc pleurodesis subset and improved for the non-talc pleurodesis subset.None of  the variables that were dependent on FDG tumour volume (TLG and MTV) were found to have significance.
Variables measuring only metabolic activity (SUVMax, SUVMean) correlated with patient overall survival for FDG.

Univariate FLT survival analysis
For the whole cohort, only the percentage change in MTV was significant.For the non-talc pleurodesis subset, none of the variables was significant.

Univariate qualitative survival analysis
mRECIST did not show any significance for prediction of overall survival.For the entire cohort ECOG scores of 0 and 2 were significant but with wide confidence intervals.ECOG scores were insignificant for both the talc and nontalc pleurodesis subsets.

Multivariate survival analysis
Included in the multivariate model were all the variables with P < 0.05 for the whole cohort (FDG Baseline SUV-Mean.FDG % Change SUVMax, FDG Change SUVMax, FDG Change SUVMean, FLT % Change MTV, ECOG 0, ECOG 2).The only variable found to be significant after multivariate cox regression analysis of the whole cohort (n = 29) was percentage change in FLT MTV (P = 0.028, HR = 1.009, 95% CI [1.001, 1.017]).We also calculated the HR and CI for a 10% change in FLT MTV (P = 0.028, HR = 1.095,CI [1.010, 1.187]), which demonstrates that a 10% increase in tumour volume corresponds with a 9.5% worsening of overall survival.
Kaplan-Meier curves were generated for a percentage change in FLT MTV as it was the only significant variable on multivariate analysis.The data were separated into three subgroups.The first cut-off was for an increase in

Discussion
We have demonstrated in this pilot study that FLT PET may have a role in monitoring MPM treatment response and prognosis.We implemented novel workflows using the MiM software platform to generate 3D tumour volumes on FDG PET and FLT PET scans to extract radiomics.Our results demonstrate that a Hounsfield unit guided approach to mapping bone on CT with semi-automated workflows may isolate and subtract bone marrow uptake from FLT PET scans to derive FLT radiomics which may have prognostic significance.Qualitative variables did not show much predictive value.The three patients who were alive at the census date had overall survival times of 1026, 985 and 901 days.These were the longest overall survival times/follow-up periods within the cohort.lded values are statistically significant.The median follow-up time was therefore the same as the median overall survival time of 350 days.The sample size was determined by the availability of patients diagnosed with MPM and was small due to the low prevalence of MPM.This is a source of bias as the cohort may not be representative.There is potential bias from left censoring as overall survival time was calculated from commencement of chemotherapy rather than date of diagnosis.On multivariate Cox regression analysis percentage change in FLT MTV after one cycle of chemotherapy was the only prognostic marker which was significant for overall survival (P = 0.046).A 10% increase in FLT MTV showed 9.5% worsened odds for overall survival, which could be of clinical significance.Log-rank testing of the FLT MTV Kaplan-Meier curves resulted in rejection of the null hypothesis (P = 0.014), suggesting that there is a statistically significant difference between the curves, and therefore a larger decrease in FLT MTV correlates with increased overall survival time.We acknowledge that our methodology for choosing the cut-offs for the curves is arbitrary due to a lack of established literature observing this effect, and that other cut-off values for different curves may have resulted in a different outcome.
Changes in FLT intensity variables (SUVmax, SUVmean and TLP) did not demonstrate the same prognostic significance as FLT tumour volume.This might suggest that the burden of tumour size rather than proliferation rate is a better prognostic marker.As no FDG variables were significant on multivariate analysis this suggests FLT may be better.
For FDG PET, changes in intensity values SUVmax and SUVmean were prognostically significant on univariate analysis, and volume dependent variables MTV and TLG were not.Increases in FDG SUVMax and SUVMean had HR <1, indicating that an increase in FDG activity after one chemotherapy cycle is protective.This suggests the degree of upregulation in metabolic activity secondary to treatment may have greater prognostic significance than burden of disease as measured by tumour volume for FDG.
We postulate that this could be due to an acute inflammatory response to the chemotherapy, and the magnitude of the acute inflammatory response may indicate how well the tumour will respond to treatment.We acknowledge that the small sample size and the short interval after completing chemotherapy in which scans were taken may be confounding these results, and an increase in FDG metabolic activity may instead indicate cancer cells resistance to treatment and disease progression.
The result of this study supports literature findings that talc pleurodesis confounds FDG PET radiomics.Kwek et al. demonstrated a correlation between pleural thickening on CT from previous talc pleurodesis and increased FDG PET avidity. 14Nowak et al. demonstrated in mesothelioma patients with prior talc pleurodesis that it confounds TLG's prognostic value. 7Table 2 demonstrates how the statistical significance of FDG metabolic variables between cycles (SUVMax change, SUVMean change) improved across the non-talc pleurodesis group relative to the talc pleurodesis group, and did not remain significant for the whole cohort on multivariate analysis.The small and differing sample sizes of each subset must be considered when observing this trend and may be influencing the P-values.
While a percentage change in FLT MTV was the only significant variable for the whole cohort on multivariate analysis, the data do not support our hypothesis that FLT PET would not be confounded by talc pleurodesis.Both the non-talc (P = 0.080) and talc pleurodesis (P = 0.208) subgroups were statistically insignificant for a percentage change in FLT MTV.We therefore cannot conclude that FLT PET is not confounded by prior talc pleurodesis.It is possible that the small sample sizes of the subgroups has contributed to their statistical insignificance.For future studies, we would recommend further testing of FLT MTV as a prognostic marker for mesothelioma with large multicentre trials to generate larger cohorts to better compare these two subgroups.
Generation of our FLT MTVs required transposing of the FDG MTVs onto the FLT images.It is therefore possible that the talc pleurodesis captured in the FDG MTVs could have potentially confounded the FLT MTV radiomics.We attempted to mitigate this by adjusting the MTVs once transposed onto the FLT images by removing areas of the volume that did not correspond to areas of increased FLT activity.We acknowledge that this process itself will be subject to interobserver variability.While our data suggest that a Hounsfield guided approach for the removal of bone marrow on FLT may improve FLT accuracy and make it a viable imaging modality, we postulate that as automated region growing software improves, techniques for extracting FLT MTVs will also improve and increase data accuracy by reducing the need for manual adjustments.
A limitation of this study was that FLT and FDG PET scans were only acquired from patients at baseline and after one cycle of chemotherapy.The ideal timing of PET imaging post chemotherapy in mesothelioma has not been established.As mesothelioma is not highly chemo sensitive, comparison between baseline scans and scans after two cycles of chemotherapy may be a more appropriate time-point to discriminate responders from non-responders.Scans were also taken 4 days after completion of a chemotherapy cycle.This may have been too short an interval for changes in response to the treatment to adequately manifest on imaging, potentially reducing the observed magnitude of change in tumour volume and markers of metabolic activity and proliferation.
By imaging after only one cycle of chemotherapy, the ability of FLT or FDG PET to detect a significant change may have been reduced.FDG scans were always obtained before FLT scans, and there may be bias from disease progression in the interval.There are no secular trends with regards to MPM treatment which would have affected patient outcomes and management over the trial period that could have caused bias.We acknowledge the small sample size of the study and recommend that future multicentre studies recruit cohorts large enough to attempt to validate these results, although this would be difficult given low disease prevalence.
The radiomic analysis performed in this study is limited to assessing tumour metabolic activity and proliferation over the entire derived tumour mass and not at the level of individual voxels.Our radiomics assume homogeneity in the structure of the tumour, which is an over-simplification in MPM, as there is heterogeneity in both FDG and FLT activity throughout the tumour.We postulate that a voxel-based analysis which assesses changes in each voxel within the tumour after chemotherapy be more indicative of tumour activity than the current method which treats the entire tumour as a single unit from which to derive radiomics.The workflows developed using MiM could be expanded upon to perform this more complex analysis.
This study is the first published study which prospectively evaluates FLT PET in patients with mesothelioma, both at baseline and after chemotherapy.The results suggest that FLT PET MTV may better predict overall survival than FDG PET, but we cannot conclude that FLT PET is not confounded by prior talc pleurodesis as it did not show statistical significance for both the non-talc and talc pleurodesis subgroups.
In conclusion, this is the first study demonstrating that radiomics extracted from tumour volumes generated on FLT PET may be predictive for assessing early treatment response of malignant pleural mesothelioma to platinumbased chemotherapy.An increase in FLT metabolic tumour volume was the only statistically significant variable on multivariate analysis for predicting patient overall survival.Talc pleurodesis was found to confound FDG PET metabolic parameters (changes in SUVmax and SUVmean), and they were not significant on multivariate analysis.We were unable to demonstrate that FLT PET is not confounded by talc pleurodesis.The statistical significance of the FLT PET volumes suggests Hounsfield guided CT bone mapping may be a viable way to isolate FLT bone marrow uptake.

Fig. 1 .
Fig. 1.FDG and FLT PET scans showing mesothelioma extending from the left lung apex into the left supraclavicular fossa.(a) FDG PET at baseline.Note the physiological cardiac ventricular uptake on FDG.(b) FDG PET after one chemotherapy cycle.(c) FLT PET at baseline.(d) FLT PET after one chemotherapy cycle.Note the bone marrow activity on FLT.

Fig. 2 .
Fig. 2. FLT PET workflow using MIM.(a) FDG Volume.(b) FDG Volume transposed onto FLT.(c) CT mapped bone volumes mapped onto (b).(d) Remaining volume on FLT after subtracting bone volumes and further refining with MiM's PET Edge tool.

Figure 4
Figure 4 illustrates patient flow from recruitment into talc pleurodesis and non-talc subsets.On follow-up 26 were deceased, and three survived by the census date.One patient had an unrecorded ECOG score, and thus ECOG P-values were calculated with n = 28.The only clinical interventions were scheduled chemotherapy sessions as detailed under methods.All 29 patients had MTVs generated from both FDG and FLT PET scans.Table1summarises cohort demographics, with CT Stage from the start of treatment.Median overall survival time was 350 days, and the shortest was 29 days.Tables2 and 3summarise the P-values, hazard ratios, and 95% confidence intervals for quantitative FDG and FLT variables, and qualitative variables respectively.

FLT
MTV.The median value for a decrease in FLT MTV by percentage change was 18.48%.The second group cutoff was therefore set at an 18% decrease in FLT MTV.The three data sub-groups were an FLT MTV decrease >18% (n = 11), FLT MTV decrease <18% (n = 10), and increased MTV (n = 8).

Fig. 5 .
Fig. 5. Kaplan-Meier curve illustrating the correlation between percentage change in FLT MTV between baseline and after one cycle of chemotherapy with overall survival time.

Table 1 .
Patient demographics and quantitative variables patient had ECOG score not recorded; hence, n = 28 for ECOG.© 2023 The Authors.Journal of Medical Imaging and Radiation Oncology published by John Wiley & Sons Australia, Ltd on behalf of Royal Australian and New Zealand College of Radiologists.

Table 2 .
P-values, hazard ratios and 95% confidence intervals for quantitative variables.
© 2023 The Authors.Journal of Medical Imaging and Radiation Oncology published by John Wiley & Sons Australia, Ltd on behalf of Royal Australian and New Zealand College of Radiologists.

Table 3 .
P-values hazard ratios, and 95% confidence intervals for qualitative variables.Bolded values are statistically significant