Predicting pathological complete response (pCR) after stereotactic ablative radiation therapy (SABR) of lung cancer using quantitative dynamic [¹⁸F]FDG PET and CT perfusion: a prospective exploratory clinical study

Stereotactic ablative radiation therapy (SABR) is effective in treating inoperable stage I non-small cell lung cancer (NSCLC) [1]. The majority of studies of using SABR for this indication report a 3-year local control rate without surgery of approximately 90%, based on imaging [2]. However, tumour response after SABR is difficult to evaluate using current clinical evaluation guidelines, such as RECIST (Response Evaluation Criteria in Solid Tumours version 1.1) and PERCIST (PET Response Criteria in Solid Tumours version 1.0) guidelines.

After SABR, more than 50% of patients can have lung density changes on computed tomography (CT) imaging which could be due to radiation-induced lung injury (RILI) or tumour recurrence [3]. However, these changes do not have distinctive patterns that easily differentiate between RILI and recurrence. Moreover, the inflammatory RILI response shows a high fluorine-18 fluorodeoxyglucose ([¹⁸F]FDG) standardized uptake value (SUV) on positron emission tomography (PET) imaging, implying a hypermetabolic state as in tumour recurrence [4]. Because RILI and recurrence can have similar size, morphology and [¹⁸F]FDG uptake [5, 6], differentiating these entities based on post-SABR CT and PET according to RECIST and PERCIST criteria can be difficult.

CT Perfusion (CTP) and dynamic [¹⁸F]FDG-PET (as opposed to static SUV measurement taken at one time point) can be used to understand tissue hemodynamics (blood flow (delivery) and permeation of the endothelial barrier) and glucose metabolism, respectively. The two modalities have a practical advantage that they can be conveniently combined together in a single study session on a PET/CT scanner. Evaluation of perfusion, blood volume, and glucose uptake rate could be more sensitive than either RECIST or PERSIST at monitoring the response to SABR. The purpose of the present study was to evaluate whether the imaging biomarkers from CTP and dynamic [¹⁸F]FDG-PET could predict the true pathologic complete response (pCR) of NSCLC to SABR. To our knowledge, this is the first prospective study evaluating imaging-based biomarkers to predict pCR after SABR treatment with histopathological study on the explanted tumour as the gold standard.

Between September 2014 and September 2017, 40 patients were enrolled in this study, of which 35 were evaluable for the primary endpoint, as described in the original paper [7]. Of those 35, 26 patients completed both imaging sessions and were available to be analyzed. The other 9 declined or was unavailable for one or both of the imaging sessions. Patient enrollment is summarized in Fig. 1. Table 1 shows the clinical-pathologic characteristics of the patients. There were 13 male and 13 female patients in this sub-study. Of the 26 patients included herein, 13 patients had RILI and all had a pCR, and 13 patients had residual disease.

The RPA identified three patient groups based on tumour blood volume before SABR (BV_pre-SABR) and change in SUV_max (ΔSUV_max) as shown in Fig. 2. According to the RPA, group 1 was defined as baseline tumour blood volume (BV_pre-SABR) ≥ 9.3 mL/100 g (n = 6, 0% pCR rate). No Group 1 patient showed pCR from the pathological analysis. All 6 patients had tumour residual disease after SABR treatment. Group 2 was defined by BV_pre-SABR < 9.3 mL/100 g and the percent change in SUV_max (ΔSUV_max) ≥ − 48.9% after SABR (n = 8, 25% pCR rate). Group 3 was defined by BV_pre-SABR < 9.3 mL/100 g and ΔSUV_max < − 48.9% after SABR (n = 12, 92% pCR rate).

The RPA model was able to predict pCR with 85% sensitivity, 92% specificity, 92% PPV, 86% NPV, and concordance (area under the ROC curve) of 0.92 (95% confidence interval (CI) 0.82–1.00). In contrast, current clinical RECIST criteria of CR/PR applied to pre- and post-SABR from the same patient showed 46% sensitivity, 47% specificity, 38% PPV, 54% NPV, and concordance of 0.54 (95% CI 0.34–0.74) in predicting pCR. Furthermore, PERCIST criteria of CMR/PMR applied to pre- and post-SABR [¹⁸F]FDG scan showed 85% sensitivity, 31% specificity, 55% PPV, 67% NPV, and concordance of 0.58 (95% CI 0.35–0.80) in predicting pCR. DeLong test of the C-statistic showed that pCR prediction with RPA model was significantly different from RECIST and PERCIST (p < 0.01). RECIST vs. PERCIST for pCR prediction was not different (p = 0.81). ROC curves are shown in Fig. 3. Figure 4 shows increase in lesion diameter on CT and decrease in uptake of [¹⁸F]FDG in a patient with pCR while Fig. 5 shows the opposite in another patient with pCR-decreased diameter on CT and increased [¹⁸F]FDG uptake.

This study employed RPA decision trees to predict pCR after SABR in patients with early-stage NSCLC using biomarkers from [¹⁸F]FDG-PET and CTP. Our results suggest that pCR of early-stage NSCLC to SABR can be predicted with 85% sensitivity, 92% specificity, 92% PPV, and 86% NPV using biomarkers from [¹⁸F]FDG-PET and CTP study before and 8–10 weeks after treatment.

In comparison, RECIST and PERCIST criteria showed worse pCR prediction due to radiation induced lung injury (RILI) because recurrent tumour can show similar size and morphology change and [¹⁸F]FDG uptake as RILI. Examples from Figs. 4 and 5 may indicate why the RECIST and PERCIST criteria lack sensitivity, specificity, PPV and NPV in predicting pCR to SABR. To be effective in treating early-stage NSCLC, SABR requires better evaluation criteria for response than the current RECIST and/or PERCIST; our study suggests that combined [¹⁸F]FDG and CTP could be a viable alternative.

This study suggests that pCR to SABR could be predicted with the BV of NSCLC before and change in maximum uptake of [¹⁸F]FDG post-treatment. The predictive model using RPA decision trees separated SABR patients into three groups: Group 1: patients with BV_pre-SABR ≥ 9.3 mL/100 g; Group 2: patients with BV_pre-SABR < 9.3 mL/100 g and ΔSUV_max ≥ − 48.9%; and Group 3: BV_pre-SABR < 9.3 mL/100 g and ΔSUV_max < − 48.9%. In a previous study with NSCLC patients, [¹⁸F]FDG SUV_max and BV of NSCLC correlated with cell proliferation marker, Ki67 and with microvessel density marker, CD34 staining, respectively [11]. While angiogenesis is associated with microvessel density, it does not necessarily lead to high blood flow (delivery) because the increased interstitial fluid pressure (edema) from the immature and leaky neo-vessels may reduce blood flow, resulting in tissue hypoxia [12]. Therefore, our results are consistent with the notion that high BV may induce hypoxia thereby worsen the outcome of SABR while decrease in SUV_max is a key factor to predict the treatment outcome as it reflects slowdown of tumour proliferation.

Our study has several limitations. First was the small sample size. To address this limitation, we used RPA to identify imaging-based biomarkers that have potential for predicting pCR rather than the conventional logistic regression technique which requires larger sample sizes to improve model stability. [13, 14] Second limitation is that the PET image analysis was done without image registration to minimize blur from the motion of the tumour. Third limitations is that CT and PET for RECIST and PERCIST were not done with standard clinical protocol. CT was the average map of a CTP study done without breathhold, even though the images were co-registered before they were averaged together, there may still be residual motion blur. The PET SUV scan was acquired at 30 min post injection, the clinical SUV scan is normally acquired between 50 and 70 min post-injection. Fourth limitation is that pCR in some tumours might have been missed because the interval time (2 weeks) between completion of SABR and surgery could be too short. Nevertheless, we were able to create a predictive model based on imaging biomarkers from CTP and PET study that was significantly better than RECIST and PERCIST criteria; therefore, this RPA model warrants future external validation with larger sample size studies.

In conclusion, this study shows that tumour BV before treatment (BV_pre-SABR) and change in [¹⁸F]FDG SUV_max (ΔSUV_max) at 8 weeks post-treatment can predict pCR of early-stage NSCLC to SABR with good sensitivity and high specificity. In comparison, RECIST and PERCIST criteria had poorer sensitivity and specificity in pCR prediction. While these findings were limited by the small sample size, the developed prediction model warrants further investigation.