Pre-treatment prediction of early response to chemoradiotherapy by quantitative analysis of baseline staging FDG-PET/CT and MRI in locally advanced cervical cancer

Abstract

Background

Early prediction of response to concurrent chemoradiotherapy (cCRT) could aid to further optimize treatment regimens for locally advanced cervical cancer (LACC) in the future.

Purpose

To explore whether quantitative parameters from baseline (pre-therapy) magnetic resonance imaging (MRI) and FDG-PET/computed tomography (CT) have potential as predictors of early response to cCRT.

Material and Methods

Forty-six patients with LACC undergoing cCRT after staging with FDG-PET/CT and MRI were retrospectively analyzed. Primary tumor volumes were delineated on FDG-PET/CT, T2-weighted (T2W)-MRI and diffusion-weighted MRI (DWI) to extract the following quantitative parameters: T2W volume; T2W signal_mean; DWI volume; ADC_mean; ADC_SD; MTV_42%; and SUV_max. Outcome was the early treatment response, defined as the residual tumor volume on MRI 3–4 weeks after start of external beam radiotherapy with chemotherapy (before the start of brachytherapy): patients with a residual tumor volume <10 cm³ were classified as early responders. Imaging parameters were analyzed together with FIGO stage to assess their performance to predict early response, using multivariable logistic regression analysis with bi-directional variable selection. Leave-one-out cross-validation with bootstrapping was used to simulate performance in a new, independent dataset.

Results

T2W volume (OR 0.94, P = 0.003) and SUV_max (OR 1.15, P = 0.18) were identified as independent predictors in multivariable analysis, rendering a model with an AUC of 0.82 in the original dataset, and AUC of 0.68 (95% CI 0.41–0.81) from cross-validation.

Conclusion

Although the predictive performance achieved in this small exploratory dataset was limited, these preliminary data suggest that parameters from baseline MRI and FDG-PET/CT (in particular pre-therapy tumor volume) may contribute to prediction of early response to cCRT in cervical cancer.

Keywords

Cervix primary neoplasms positron emission tomography magnetic resonance imaging magnetic resonance diffusion/perfusion

Introduction

In cervical cancer, disease stage is typically determined at diagnosis by a combination of clinical examination and pelvic magnetic resonance imaging (MRI), complemented with whole-body 18F-fluoro-deoxyglucose (FDG) positron emission tomography/computed tomography (PET/CT) for evaluation of lymph nodes and distant metastases in more advanced cases. Current treatment standard for locally advanced cervical cancer (LACC; International Federation of Gynecology and Obstetrics [FIGO] stage ≥IB2 or node-positive) is definitive concurrent chemoradiotherapy (cCRT), consisting of weekly cisplatin in combination with external beam radiotherapy (EBRT), followed by brachytherapy (1).

To date, the radiotherapy schedule is generally identical for all patients, irrespective of tumor stage or other prognostic characteristics. In the vast majority, this regime results in a complete local response (2), but there is considerable variation in the course of response between individual patients. In some patients, a significant tumor residue remains after the first weeks of EBRT (before brachytherapy), while others are “early responders” that already show a (near-)complete volume reduction at this stage (3). In the future, these “early responders” might benefit from an early start of the subsequent brachytherapy, aiming to reduce overall treatment time, or from dose de-escalation to limit treatment toxicity while maintaining a good oncological outcome (4). To facilitate such personalized treatments, tools to predict early response will be required.

One approach could be to extract such predictive information from imaging. Studies in various cancer types have shown promising results for imaging biomarkers derived from MRI and FDG-PET/CT as predictors for response and prognosis (5 –8). For cervical cancer, there have been reports that markers from pre-treatment MRI or FDG-PET/CT have potential to predict “late” outcomes, including disease-free and overall survival (9 –16) or final treatment response after completion of cCRT (17 –26); however, so far, no studies have focused specically on prediction of early treatment response. Furthermore, of the existing studies, only a few combined FDG-PET/CT and MRI within one patient cohort (9,12,18,27).

The aim of the present study was to determine whether quantitative imaging markers from baseline staging MRI and FDG-PET/CT have potential as predictors of early response to cCRT in LACC, and whether combining parameters from both modalities has complementary value.

Material and Methods

This retrospective study was approved by the institutional ethical review board, informed consent was waived. The study workflow is illustrated in Fig. 1.

Fig. 1.

Study workflow. ^aMRI acquired during cCRT for clinical brachytherapy planning. For the purpose of this study, this MRI was used for volumetric response measurement. cCRT, concurrent chemoradiotherapy; MRI, magnetic resonance imaging.

Patients and treatment

Patients who underwent FDG-PET/CT and MRI in our institution for pre-treatment staging of primary cervical cancer (January 2011 to March 2018) were identified. From this group, 46 patients met the inclusion criteria: (i) LACC (FIGO stage ≥IB2 or node-positive (28)); (ii) treatment by cCRT (with curative intent) and at least four of six weekly cycles of cisplatin completed; (iii) pre-treatment MRI at 3.0 T including T2-weighted (T2W) and diffusion-weighted imaging (DWI); and (iv) available MRI at 3–4 weeks after EBRT initiation, to determine “early response.”

Routine cCRT consisted of pelvic EBRT of 46 Gy (2 Gy/fraction, five fractions/week). The radiation field was extended to the level of the renal veins in cases of suspicious para-aortic nodes. A sequential boost (14 Gy, 2 Gy/fraction, five fractions/week) was administered to suspected sites of pelvic lymph-node involvement. EBRT was followed by brachytherapy to a total dose equivalent of 90 Gy on the high-risk clinical target volume (cervix and tumor), in 3–4 fractions. Radiotherapy was accompanied by weekly cisplatin (40 mg/m² body surface area) for six weeks, starting on day 1 of EBRT.

MRI

Pre-therapy pelvic MRI was performed at 3.0 T (Intera Achieva [+/− dStream] or Ingenia system, Philips Healthcare) with an external surface coil. The protocol included anatomical fast spin echo (FSE) T2W sequences in three orthogonal planes and an axial single-shot echo planar imaging (ssEPI) DWI sequence, with three b-values (b 0 up to b 750–1000). Axial T2W and DWI sequences were angled in identical planes (perpendicular to the cervical canal). Protocol details are provided in Table 1. Spasmolytics were not administered.

Table 1.

MRI protocol used for primary staging and quantitative imaging evaluation.

	T2W FSE	Single-shot EPI DWI
Repetition time (ms)	2515.84–4865.10	1200.0–8824.4
Echo time (ms)	110–150	54.03–97.97
Slice thickness (mm)	3–4	2.7–3
Slice gap (mm)	3–4	3.2–3.5
In-plane resolution (mm)	0.25 × 0.25–0.49 × 0.49	0.56 × 0.56–1.48 × 1.48
Echo train length	25–28	67–113
No. signal averages	1–2	1–5
b-values (used for calculation of ADC) (s/mm²)	–	0, 200, 800*
Fat suppression	–	SPAIR

*Some changes in protocol and sequence parameters occurred during the study period: in 18 patients, ADC was calculated using slightly different b-values: 0, 188, 750. In two patients, ADC was calculated using b-values 0, 200, 1000.

ADC, apparent diffusion coefficient; DWI, diffusion-weighted imaging; EPI, echo planar imaging; FSE, fast spin echo; MRI, magnetic resonance imaging; SPAIR, spectral attenuated inversion recovery; T2W, T2-weighted.

FDG-PET/CT imaging

FDG-PET/CTs were performed on a hybrid PET/CT scanner (either Gemini TF 16 or Gemini TF Big Bore, Philips Healthcare). After 6 h of fasting (target blood glucose level <10 mmol/L), patients received an intravenous 2-deoxy-2-[18F]fluoro-D-glucose (FDG) bolus of 180 MBq (for body mass index [BMI] ≤28 kg/m²) or 240 MBq (BMI >28 kg/m²), followed by an accumulation period of 60 ± 5 min. Scanning ran from the skull base to upper thighs, with 2 min/bed position (reconstructed to 4-mm slices, 4 × 4 mm pixels), with a non-enhanced CT (120–140 kV, target energy 40 mAs with automatic dose modulation) for attenuation correction and anatomical correlation (reconstructed to 2 mm and 5 mm slices, 1.17 × 1.17 mm pixels).

Tumor segmentation

A board-certified abdominal radiologist (DL, with ±7 years of pelvic MRI experience) manually segmented whole-tumor volumes on the primary staging axial T2W-MRI and subsequently (in the same session) on high b-value DWI, using dedicated open-source software (3D Slicer, version 4.8.1). To assess inter-observer variation, a random subset of 15 cases was additionally segmented by a second board-certified radiologist (MM, with similar experience level), blinded for the results of the first reader.

FDG-PET/CT images were processed by a nuclear medicine physician (WV, with 13 years of experience), using the same software. The FDG-avid tumor volume was first manually segmented, while carefully excluding adjacent structures with high physiological FDG signal (e.g. urine in the bladder). The metabolic tumor volume was derived semi-automatically from this segmentation using a threshold of 42% of the SUV_max (MTV_42%, according to methods previously reported (11,13,29)).

Quantitative image analysis

Using the open-source software PyRadiomics (version 2.1.0 (30)) the following quantitative parameters were extracted from the segmented volumes: T2W volume; DWI volume; mean T2W signal (T2W-signal_mean); mean apparent diffusion coefficient (ADC_mean); the SD of the ADC (ADC_SD); SUV_max; and MTV_42%. These specific parameters were selected as they represent relatively simple (first-order) parameters reflecting tumor size, cellularity, heterogeneity, and metabolism, all of which have shown potential for response prediction in previous studies (10,16,17,24). ADC values were calculated using a mono-exponential decay model including three b-values (b = 0, b = 88/200, b = 750/800/1000 s/mm²). MRI data were resampled to account for variation in the voxel dimensions, and signal intensity was normalized (by subtracting the image mean value and dividing it by the SD).

Outcome definition (early response)

The primary study outcome was the early response to cCRT, defined as the residual tumor volume on MRI (routinely performed for the purpose of brachytherapy planning) after 3–4 weeks of cCRT. This outcome was chosen as the residual tumor volume after EBRT but before brachytherapy has previously been described to correlate with final local control after completion of treatment (19,31,32). The volume threshold was set at <10 cm³, based on a previously reported cut-off (19). The tumor volume after EBRT was segmented on T2W-MRI by the same reader and using the same procedure and segmentation methods as for the initial tumor segmentations. In 29/46 patients, response evaluation was performed on a 1.5-T scanner (Intera Achieva, Philips Healthcare) using similar protocols as the standard 3.0-T protocol described above and in Table 1.

Statistical analysis

Inter-observer variability was determined with the intraclass coefficient (ICC) in a two-way random effects model. The predictive value of the different imaging parameters (T2W volume, T2W signal_mean, DWI volume, ADC_mean, ADC_SD, SUV_max, MTV_42%), together with clinical FIGO stage, to predict early response to cCRT was assessed using multivariable logistic regression analysis with bi-directional stepwise selection based on the Akaike Information Criterion (AIC) (33). FIGO stage was derived from the patients’ electronical medical records (according to the 2009 FIGO staging system, routinely used during the inclusion period of the present study) (28). In case of strong correlation between different imaging parameters (Pearson’s correlation coefficient ρ ≥ 0.8), only one of these parameters was used in the variable selection process to avoid effects of multicollinearity. Model performance was determined by calculating the area under the receiver operating characteristic (ROC) curve (AUC), and odds ratios (ORs) were calculated for all parameters selected in the model. To assess model performance in new, “independent” data, leave-one-out cross-validation (LOOCV) was performed. For LOOCV, each patient is left out of the dataset once and a prediction model is generated on the remaining cases (46 iterations for this study). Probability of early response is calculated by the model for the left-out patient. The cross-validated AUC is determined on the total of these predictions. A 95% bootstrap confidence interval for the cross-validated AUC was obtained by repeating the cross-validation process using 1000 bootstrapped samples. Univariable logistic regression analysis for the individual variables was performed independent of the multivariable analysis. Baseline variables were compared between outcome groups using Wilcoxon Mann–Whitney U test for independent samples or Fisher’s exact test.

Statistical analyses were performed using R version 3.4.3 (R Foundation for Statistical Computing, Vienna, Austria).

Results

Baseline characteristics

Baseline patient characteristics are given in Table 2. The median age was 51 years (age range = 27–75 years). Most patients had FIGO stage IIB: 34/46 (74%). In total 30/46 (65%) patients were classified as early responders (residual T2W volume <10 cm³). Total treatment time (EBRT + brachytherapy) was in the range of 5.3–8.3 weeks (median = 6.3 weeks).

Table 2.

Baseline characteristics of included patients (n = 46).

		Residual T2W volume <10 cm³ (good early response)	Residual T2W volume ≥10 cm³	P value*
Patients (n)		30	16	n/a
Age (years)		49 (28–61)	52 (27–75)	0.188
FIGO stage^†	IB2	2 (6.7)	1 (6.3)
	IIA–IIB	23 (76.7)	11 (68.8)
	IIIA–IIIB	4 (13.3)	3 (18.8)
	IV	1 (3)	1 (6.3)	0.549
Node-positive disease		20 (66.7)	13 (81.3)	0.493^‡
Tumor histology	Squamous cell carcinoma	27 (90)	11 (68.8)
	Adenocarcinoma	1 (3)	4 (25)
	Other (adenosquamous, undefined)	2 (6)	1 (6.3)	0.079^‡
Median interval between startof EBRT and second MRI (days)		23 (16–25)	23 (16–24)	0.228

Values are given as n (%) or median (range).

*P > 0.05 was considered statistically significant.

^†Stage according to the FIGO staging system of 2009.

^‡Fisher’s exact test, others: Wilcoxon Mann–Whitney U test (independent samples).

EBRT, external beam radiotherapy; FIGO, International Federation of Gynecology and Obstetrics; MRI, magnetic resonance imaging; T2W, T2-weighted.

Inter-observer variability

Inter-observer agreement was excellent for all quantitative image parameters, with ICCs in the range of 0.916–0.995 (Table 3).

Table 3.

ICC for MRI parameters.

Variable	ICC (95% CI)
T2W volume	0.972 (0.920–0.990)
DWI volume	0.916 (0.772–0.971)
T2W-signal_mean	0.995 (0.985–0.998)
ADC_mean	0.992 (0.976–0.997)
ADC_SD	0.944 (0.845–0.981)
Post-EBRT T2W_volume	0.932 (0.769–0.978)

ADC, apparent diffusion coefficient; CI, confidence interval; DWI, diffusion-weighted imaging; EBRT, external beam radiotherapy; ICC, intraclass correlation coefficient; MRI, magnetic resonance imaging; T2W, T2-weighted.

Univariable analysis

Results of the univariable analysis are provided in Table 4. Among the quantitative imaging parameters, all parameters related to the pre-therapy primary tumor volume (T2W volume, DWI volume, and MTV_42%) were significantly associated with early response, with ORs in the range of 0.94–0.95 (per cm³), indicating that a smaller tumor volume is associated with early response to cCRT. Of the other parameters, only ADC_SD and T2W-signal_mean showed borderline-significant associations (OR = 1.06, P = 0.07 and OR = 0.98, P = 0.09, respectively). FIGO stage category was not significantly associated with early response (OR = 0.60, P = 0.50).

Table 4.

Univariable association with early response.

Variable	OR (95% CI)	P value
FIGO stage (III + IV vs. I + II)	0.60 (0.14–2.65)	0.50
T2W volume (cm³)	0.94 (0.91–0.98)	<0.01*
T2W-signal_mean (unit; normalized)	0.98 (0.96–1.00)	0.09
DWI volume (cm³)	0.95 (0.91–0.98)	<0.01*
ADC_mean (unit; normalized)	1.01 (0.99–1.04)	0.38
ADC_SD (unit; normalized)	1.06 (1.00–1.12)	0.07
MTV_42% (cm³)	0.94 (0.90–0.98)	<0.01*
SUV_max (unit)	0.96 (0.86–1.07)	0.44

*P values <0.05 were considered statistically significant.

ADC, apparent diffusion coefficient; CI, confidence interval; DWI, diffusion-weighted imaging; EBRT, external beam radiotherapy; FIGO, International Federation of Gynecology and Obstetrics; MTV, metabolic tumor volume; OR, odds ratio; T2W, T2-weighted.

Multivariable prediction model

Multivariable analysis was performed independently of the univariate results. Since a strong correlation (ρ = 0.8–0.9, P <0.01) was found between the three volume parameters (T2W volume, DWI volume, and MTV_42%), T2W volume was chosen as the only volume-based parameter entered in the variable selection process because it corresponds best with the outcome definition (volume on T2W-MRI after EBRT). Results are summarized in Table 5. The multivariate prediction model included T2W volume (OR = 0.94, P = 0.003) and SUV_max (OR = 1.15, P = 0.18) as the selected predictors. Median pre-CRT T2W volume was 24.8 cm³ (range = 7.2–53.7 cm³) for the early responders versus 64.1 cm³ (range = 21.2–134.9 cm³) for the remaining patients. Median SUV_max for the early responders was 15.3 (range = 6.7–26.1) versus 15.9 (range = 4.8–33.6) for the remaining patients. The model’s performance to predict early response was AUC 0.82 within the current dataset. AUC by cross-validation (LOOCV) was 0.68 (95% CI = 0.41–0.81).

Table 5.

Multivariable logistic prediction model.

Variable	OR (95% CI)	P value
T2W volume (cm³)*	0.93 (0.89–0.98)	<0.01
SUV_max	1.15 (0.94–1.40)	0.18
AUC	0.82
AUC LOOCV^†	0.68 (0.41–0.81)

*There was a strong correlation (ρ = 0.8–0.9, P < 0.01) between T2W volume, DWI volume, and MTV_42%. Therefore, only T2W volume was used in the variable selection process.

^†LOOCV with bootstrapping was used to account for overfitting and to estimate how the model would perform on new, independent data.

ADC, apparent diffusion coefficient; AUC, area under the receiver operating characteristic curve; CI, confidence interval; LOOCV, leave-one-out cross-validation; OR, odds ratio; T2W, T2-weighted.

Discussion

Previous studies have shown that pre-therapy quantitative imaging markers from FDG-PET/CT and MRI have potential to predict “late” outcomes including the final treatment response (17 –22,24 –26) and survival (9 –15) in patients with LACC. In addition to these previous works, the present study specifically assessed whether quantitative imaging parameters from baseline staging MRI (including DWI) and FDG-PET/CT may also have potential as predictors of early response to cCRT and could thus play a possible role in future optimization of treatment regimens. Our exploratory results indicate that the pre-therapy tumor volume is the best predictor (with similar results for volumes derived from T2W-MRI, DWI, and FDG-PET/CT in univariable analysis) for early response to EBRT (i.e. response before onset of brachytherapy), with smaller baseline volumes (median 24.8 vs. 64.1 cm³) for the early responders. Of the other parameters, only SUV_max may have some complementary value, though the SUV_max values between both outcome groups overlapped considerably. FIGO stage was not identified as a significant predictor for early response, nor were any of the DWI-related parameters (ADC_mean and ADC_SD). Overall, the estimated predictive performance achieved with cross-validation was only moderate (AUC ∼ 0.7) and ORs of the selected parameters were in the range of 0.94–1.15, indicating a limited predictive value, which is typically not considered sufficient for clinical decision-making. The present study was exploratory, however, and further and more in-depth research is obviously required to investigate if other (combinations of) imaging and clinical parameters yield higher predictive potential.

It is difficult to directly compare our current results with those reported by previous authors, because their studies were aimed at later endpoints (disease-free and overall survival) or final response after cCRT, while the current study focuses specifically on predicting early response to EBRT. Furthermore, few studies have combined MRI and FDG-PET/CT parameters for prediction of outcome in cervical cancer. Studies that did, in most cases found FDG-PET-based pre-therapy parameters were associated with the final outcome, including a smaller baseline metabolic tumor volume (11) and higher SUV_max (18) corresponding to a favourable outcome. In contrast, high pre-therapy SUV_max was reported to predict a poor final treatment response by Kidd et al. (16,22) looking at PET parameters only, yet others were unable to confirm an association (10,34). In three MRI-based studies that included the tumor volume on pre-treatment T2W-MRI to predict final treatment response, this parameter was not found to be a significant predictor (15,25,35). However, a study by Schernberg et al. (31) did show a significant association between pre-treatment tumor volumes and survival.

We were unable to detect a significant association between the DWI-derived parameters and early treatment response. Although association of ADC_SD nearly reached significance in univariable analysis, this parameter was not identified as an independent predictor in the multivariable analysis. Previous reports on ADC-based parameters to predict response to cCRT in cervical cancer have so far shown conflicting results. Some studies found low pre-treatment ADC_mean to be associated with a good final response to cCRT (17,20,24), while others found no significant results (11,15,21,25,35). The role of DWI to predict treatment response in cervical cancer thus remains unclear but appears to be relatively limited.

When tested within our original dataset, the multivariable model (including T2W volume and SUV_max as predictors) yielded an AUC of 0.82. This is likely an overestimation of its actual performance in hypothetical clinical application. Because our limited cohort size did not allow splitting of the dataset into separate training and validation sets, cross-validation of the bi-directional stepwise modelling process was used to simulate model performance on a new independent dataset (i.e. to estimate its clinical potential). This resulted in an AUC of 0.68, which is likely a more realistic approximation of the model’s actual performance. As mentioned, this is certainly not sufficient for clinical decision-making, and further optimization (and larger scale validation) will obviously be required. While FIGO stage—though specifically developed for prognostication—did not contribute to the current model, it will be worthwhile to further explore the added value of clinical factors, as well as additional histological, immunohistochemical, or genetic factors to generate stronger and more comprehensive clinical prediction models. It would also be worthwhile to see if the recently updated FIGO staging system would render different results than the 2009 FIGO staging system routinely used during the clinical inclusion period of the current study. Finally, sophisticated methods of image analysis and postprocessing such as Radiomics or deep learning, often used to assess large sets of variables, may prove to be of added value as the first available publications on these relatively novel methods have shown some promising preliminary results (9,18,36). In the present study, we consciously limited the number of variables to prevent overfitting in a small dataset and identified a small set of relatively intuitive (first-order) features based on previous works. We acknowledge, however, that with this approach other potentially valuable imaging biomarkers may have been overlooked.

The present study has some limitations, in addition to the aforementioned small patient cohort. First, some variations in MR acquisition protocols occurred over time, which are difficult to avoid during retrospective analysis of clinical data. We aimed to account for this by normalizing the MRI signal intensity (T2W and ADC). Second, a validated reference standard to classify early response during cCRT does not exist. We therefore chose the residual volume on T2W-MRI after 3–4 weeks of EBRT as a measure of early response, using a volume threshold of <10 cm³ derived from a study by Mongula et al. (19), who reported this as a cut-off that correlates with local control after brachytherapy. Finally, part of the MRIs used for response classification were acquired at 1.5 T instead of 3.0 T (29/46 patients). Given the comparable visual quality and resolution, this likely had little impact on the tumor volumes used for response classification.

In conclusion, these preliminary data suggest that parameters from baseline MRI and FDG-PET/CT (particularly primary tumor volume) may contribute to prediction of early response to cCRT in LACC, although the predictive performance achieved was limited. Future larger-scale studies are required to expand this research, by combining imaging markers with other potential predictors and by exploring more sophisticated image analysis techniques, to build prediction tools that can truly aid in further treatment personalization in cervical cancer.

Footnotes

Acknowledgements

The authors thank Prof. Dr. Michael Hauptmann, former head of the Biostatistics group of the Netherlands Cancer Institute, for his supervision of the statistical analysis and critical revision of the manuscript, and Chèrita Sombroek, physician assistant at the department of Radiation Oncology of the Netherlands Cancer Institute, for her assistance in cohort identification.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Lisa A Min

References

Cibula

Pötter

Planchamp

, et al. The European Society of Gynaecological Oncology/European Society for Radiotherapy and Oncology/European Society of Pathology guidelines for the management of patients with cervical cancer. Radiother Oncol 2018; 127:404–416.

Ranjan Datta

Stutz

Liu

, et al. Concurrent chemoradiotherapy vs. radiotherapy alone in locally advanced cervix cancer: A systematic review and meta-analysis. Gynecol Oncol 2017; 145:374–385.

Jastaniyah

Yoshida

Tanderup

, et al. A volumetric analysis of GTVD and CTVHR as defined by the GEC ESTRO recommendations in FIGO stage IIB and IIIB cervical cancer patients treated with IGABT in a prospective multicentric trial (EMBRACE). Radiother Oncol 2016; 120:404–411.

Pötter

Tanderup

Kirisits

, et al. The EMBRACE II study: The outcome and prospect of two decades of evolution within the GEC-ESTRO GYN working group and the EMBRACE studies. Clin Transl Radiat Oncol 2018; 9:48–60.

Giannini

Mazzetti

Bertotto

, et al. Predicting locally advanced rectal cancer response to neoadjuvant therapy with 18 F-FDG PET and MRI radiomics features. Eur J Nucl Med Mol Imaging 2019; 46:878–888.

van Rossum

PSN

Fried

Zhang

, et al. The value of 18F-FDG PET before and after induction chemotherapy for the early prediction of a poor pathologic response to subsequent preoperative chemoradiotherapy in oesophageal adenocarcinoma. Eur J Nucl Med Mol Imaging 2017; 44:71–80.

Yoon

Kim

Chung

, et al. Predicting neo-adjuvant chemotherapy response and progression-free survival of locally advanced breast cancer using textural features of intratumoral heterogeneity on F-18 FDG PET/CT and diffusion-weighted MR imaging. Breast J 2019; 25:373–380.

Min

Lin

Liney

, et al. A review of the predictive role of functional imaging in patients with mucosal primary head and neck cancer treated with radiation therapy. J Med Imaging Radiat Oncol 2017; 61:99–123.

Lucia

Visvikis

Desseroit

M-C

, et al. Prediction of outcome using pretreatment 18 F-FDG PET/CT and MRI radiomics in locally advanced cervical cancer treated with chemoradiotherapy. Eur J Nucl Med Mol Imaging 2018; 45:768–786.

10.

Guler

Torun

Yýldýrým

, et al. Pretreatment metabolic tumor volume and total lesion glycolysis are not independent prognosticators for locally advanced cervical cancer patients treated with chemoradiotherapy. Br J Radiol 2018; 91:20170552.

11.

Ueno

Lisbona

Tamada

, et al. Comparison of FDG PET metabolic tumour volume versus ADC histogram: Prognostic value of tumour treatment response and survival in patients with locally advanced uterine cervical cancer. Br J Radiol 2017; 90:20170035.

12.

Park

Kim

Park

BK.

Prognostic value of diffusion-weighted magnetic resonance imaging and 18 F-fluorodeoxyglucose-positron emission tomography/computed tomography after concurrent chemoradiotherapy in uterine cervical cancer. Radiother Oncol 2016; 120:507–511.

13.

Miccò

Vargas

Burger

, et al. Combined pre-treatment MRI and 18F-FDG PET/CT parameters as prognostic biomarkers in patients with cervical cancer. Eur J Radiol 2014; 83:1169–1176.

14.

Nakamura

Kajitani

Joja

, et al. The posttreatment mean apparent diffusion coefficient of primary tumor is superior to pretreatment adcmean of primary tumor as a predictor of prognosis with cervical cancer. Cancer Med 2013; 2:519–525.

15.

Somoye

Harry

Semple

, et al. Early diffusion weighted magnetic resonance imaging can predict survival in women with locally advanced cancer of the cervix treated with combined chemo-radiation. Eur Radiol 2012; 22:2319–2327.

16.

Kidd

Siegel

Dehdashti

, et al. The standardized uptake value for F-18 fluorodeoxyglucose is a sensitive predictive biomarker for cervical cancer treatment response and survival. Cancer 2007; 110:1738–1744.

17.

Yang

Qiang

Tian

, et al. Multi-parametric MRI in cervical cancer: early prediction of response to concurrent chemoradiotherapy in combination with clinical prognostic factors. Eur Radiol 2018; 28:437–445.

18.

Bowen

Yuh

WTC

Hippe

, et al. Tumor radiomic heterogeneity: Multiparametric functional imaging to characterize variability and predict response following cervical cancer radiation therapy. J Magn Reson Imaging 2018; 47:1388–1396.

19.

Mongula JE, Slangen BFM, Lambregts DMJ, et al. Consecutive magnetic resonance imaging during brachytherapy for cervical carcinoma: predictive value of volume measurements with respect to persistent disease and prognosis. Radiat Oncol 2015;10: Article No. 252.

20.

Liu

Sun

Bai

, et al. Time-window of early detection of response to concurrent chemoradiation in cervical cancer by using diffusion-weighted MR imaging: a pilot study. Radiat Oncol 2015; 10:185.

21.

Makino

Morishige

Kato

, et al. Predictive value of diffusion-weighted magnetic resonance imaging during chemoradiotherapy for uterine cervical cancer. J Obstet Gynaecol Res 2013; 40:1098–1104.

22.

Kidd

Thomas

Siegel

, et al. Changes in Cervical Cancer FDG Uptake During Chemoradiation and Association With Response Radiation Oncology. Radiat Oncol Biol 2013; 85:116–122.

23.

Vandecasteele

Delrue

Lambert

, et al. Value of magnetic resonance and 18FDG PET-CT in predicting tumor response and resectability of primary locally advanced cervical cancer after treatment with intensity-modulated arc therapy: A prospective pathology-matched study. Int J Gynecol Cancer 2012; 22:630–637.

24.

Liu

Bai

Sun

, et al. Diffusion-weighted imaging in predicting and monitoring the response of uterine cervical cancer to combined chemoradiation. Clin Radiol 2009; 64:1067–1074.

25.

Harry

Semple

Gilbert

, et al. Diffusion-weighted magnetic resonance imaging in the early detection of response to chemoradiation in cervical cancer. Gynecol Oncol 2008; 111:213–220.

26.

Mcveigh

Syed

Milosevic

, et al. Diffusion-weighted MRI in cervical cancer. Eur Radiol 2008; 18:1058–1064.

27.

Sala

Micco

Burger

, et al. Complementary prognostic value of pelvic Magnetic Resonance Imaging and whole-body fluorodeoxyglucose Positron Emission Tomography/Computed Tomography in the pretreatment assessment of patients with cervical cancer. Int J Gynecol Cancer 2015; 25:1461–1467.

28.

FIGO Committee on Gynecologic Oncology. FIGO staging for carcinoma of the vulva, cervix, and corpus uteri. Int J Gynecol Obstet 2014; 125:97–98.

29.

Erdi

Mawlawi

Larson

, et al. Segmentation of lung lesion volume by adaptive positron emission tomography image thresholding. Cancer 1997; 80:2505–9.

30.

Van Griethuysen JJM, Fedorov A, Parmar C, et al. Computational radiomics system to decode the radiographic phenotype. Cancer Res 2017;77:e104–e107.

31.

Schernberg

Bockel

Annede

, et al. Tumor Shrinkage During Chemoradiation in Locally Advanced Cervical Cancer Patients: Prognostic Significance, and Impact for Image-Guided Adaptive Brachytherapy. Int J Radiat Oncol Biol Phys 2018; 102:362–372.

32.

Minkoff

Gill

Kang

, et al. Cervical cancer outcome prediction to high-dose rate brachytherapy using quantitative magnetic resonance imaging analysis of tumor response to external beam radiotherapy. Radiother Oncol 2015; 115:78–83.

33.

Akaike

A New Look at the Statistical Model Identification. IEEE Trans Automat Contr 1974; 19:716–723.

34.

Bjurberg

Kjellén

Ohlsson

, et al. Prediction of patient outcome with 2-deoxy-2-[18F]fluoro-D-Glucose-Positron Emission Tomography Early During Radiotherapy for Locally Advanced Cervical Cancer. Int J Gynecol Cancer 2009; 19:1600–1605.

35.

Das

Chandramohan

Karunya

, et al. Role of conventional and diffusion weighted MRI in predicting treatment response after low dose radiation and chemotherapy in locally advanced carcinoma cervix. Radiother Oncol 2015; 117:288–293.

36.

Altazi

Fernandez

Zhang

, et al. Investigating multi-radiomic models for enhancing prediction power of cervical cancer treatment outcomes. Phys Medica 2018; 46:180–188.