Could early tumour volume changes assessed on morphological MRI predict the response to chemoradiation therapy in locally-advanced rectal cancer?

Clinical Radiology 73 (2018) 555e563

Contents lists available at ScienceDirect

Clinical Radiology journal homepage: www.clinicalradiologyonline.net

Could early tumour volume changes assessed on morphological MRI predict the response to chemoradiation therapy in locally-advanced rectal cancer? A. Palmisano a, b, *, A. Esposito a, b, A. Di Chiara a, b, A. Ambrosi b, P. Passoni c, N. Slim c, C. Fiorino d, L. Albarello e, N. Di Muzio c, R. Calandrino d, R. Rosati b, f, A. Del Maschio a, b, F. De Cobelli a, b a

Clinical and Experimental Radiology, Experimental Imaging Centre, San Raffaele Hospital, Milano, Italy Vita-Salute San Raffaele University, Milano, Italy c Unit of Radiotherapy, San Raffaele Hospital, Milano, Italy d Medical Physics, San Raffaele Hospital, Milano, Italy e Department of Pathology, San Raffaele Hospital, Milano, Italy f Department of Gastrointestinal Surgery, San Raffaele Hospital, Milano, Italy b

art icl e i nformat ion Article history: Received 10 March 2017 Accepted 11 January 2018

AIM: To investigate the potential role of an additional magnetic resonance imaging (MRI) examination performed during neoadjuvant chemoradiation therapy (CRT) in the prediction of pathological response in locally advanced rectal cancer (LARC). MATERIAL AND METHODS: Forty-eight consecutive patients with LARC underwent neoadjuvant CRT. MRI studies at 1.5 T, including high-resolution T2-weighted sequences that were acquired parallel and perpendicular to the main axis of the tumour were performed before (preMRI), during (midMRI), and 6e8 weeks after the end of CRT (postMRI). Cancer volumes (Vpre, Vmid, Vpost) were drawn manually and the reduction rate calculated (DVmid, DVpost). According to € del’s pathological tumour regression grade (TRG), patients were considered non-responders Ro (NR; TRG0-2), partial responders (PR; TRG3), and complete responders (CR; TRG4). Multivariate regression analysis was performed to identify the best MRI predictors of NR, PR, and CR. RESULTS: Twenty-five patients were considered PR (52%), 13 CR (27%), and 10 NR (22%). Tumour shrinkage mainly occurred shortly after CRT (DVmid: CR: 80
10% versus PR: 56
19% versus NR: 28
22%, p¼2.21016). Vmid, Vpost, DVmid, and DVpost correlated with TRG (p<0.001). At multivariate analysis, the combined assessment of Vmid and DVmid was selected as the best predictor of response to CRT, in that it distinguishes CR, PR, and NR early and accurately (81.5%). CONCLUSION: MidMRI allows final response assessment to neoadjuvant CRT earlier and better than the MRI performed after the end of CRT. MRI findings at midMRI may be useful to tailor patient treatment. Ó 2018 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.

* Guarantor and correspondent: A. Palmisano, Clinical and Experimental Radiology, Experimental Imaging Centre, San Raffaele Scientific Institute, Via Olgettina 60, 20132, Milano, Italy. Tel.: þ39 02 2643 6102; fax: þ39 02 2643 2165. E-mail address: [email protected] (A. Palmisano). https://doi.org/10.1016/j.crad.2018.01.007 0009-9260/Ó 2018 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.

556

A. Palmisano et al. / Clinical Radiology 73 (2018) 555e563

Introduction

MRI protocol and image analysis

Neoadjuvant chemoradiotherapy (CRT) for the treatment of locally-advanced rectal cancer (LARC) has progressively become the current standard of care of LARC in improving resectability and prognosis.1 CRT outcome is a crucial endpoint in LARC management, and recent developments in neoadjuvant regimens have increased the rate in downstaging to approximately 60%,2 with a complete response rate of up to 30%.3 These results opened the possibility of directing the patients towards a conservative strategy in case of complete response to CRT.4,5 Conversely, a minor, but not negligible, percentage of patients do not gain significant benefits from neoadjuvant treatment,6 resulting in unnecessary toxicity as well as in a delay in receiving the appropriate treatment. Therefore, in recent years, numerous efforts have been made to identify an early imaging biomarker of CRT response. Magnetic resonance imaging (MRI) provides a detailed depiction of rectal cancer and the surrounding structures, representing the method of choice for the local staging of rectal cancer.7,8 MRI is also routinely performed after CRT for the evaluation of response to neoadjuvant treatment and surgical planning, but it is known that CRT-induced fibrosis and inflammation may interfere with the correct assessment of residual cancer. Consequently, the reported accuracy of MRI in LARC re-staging is largely suboptimal.9,10 Even if MRI may fail in the assessment of T stage at post-CRT, the measurement of tumour volume reduction rate seems to provide additional prognostic information, possibly related to pathological tumour regression grade (TRG) and overall survival.11,12 Although, good correlation was found between the tumour volume reduction rate from pre-to post-CRT MRI and TRG, this approach provides late information and variable threshold, with a limited capability of distinguishing partial from complete response.13,14 Assuming that tumour shrinkage is expected to occur mostly in the earlier phases of CRT in responding patients, the aim of the present study was to investigate the usefulness of an additional MRI study performed during CRT for the early prediction of histopathological TRG.

All patients underwent MRI studies according to the following framework: before CRT for rectal cancer staging (preMRI), after nine fractions of CRT (midMRI) and after at least 6e8 weeks from the end of CRT (postMRI). All examinations were performed using a 1.5 T magnet (Achieva Nova, Philips Medical Systems, Best, The Netherlands) equipped with a five-channel cardiac phased-array coil, after rectal filling (50 ml) with Lumirem (Guerbet Group, Paris, France). Spasmolytic (scopolamine) intramuscular injection was performed to avoid artefacts from bowel peristalsis. All MRI studies included high-resolution turbo spin echo (TSE) para-axial T2-weighted sequence (echo time [TE]¼ 120 ms, repetition time [TR]¼5,785 ms, field of view [FOV]¼ 24024076 mm, matrix size¼512512, section thickness¼3 mm, intersection gap¼0.3 mm, echotrain length¼19, acquisition time¼3 minutes 10 seconds), sagittal T2-weighted sequence (TE¼100 ms, TR¼2,021 ms with drive equilibrium pulse, FOV¼24018152 mm, matrix size¼512512, section thickness: 3 mm, intersection gap¼0.3 mm, echotrain length¼21, acquisition time¼3 minutes 48 seconds), and paracoronal T2-weighted sequences (TE¼100 ms, TR¼4,421 ms, FOV¼24020072 mm, matrix size¼512512, section thickness¼3.3 mm, intersection gap¼0.3 mm, echotrain length¼14, acquisition time¼3 minutes 8 seconds), respectively perpendicular and parallel to the major tumour axis. At each time point, the tumour volume was manually outlined by a radiologist with 8 years’ experience in rectal cancer using a para-axial high-resolution T2-weighted sequence using Olea Sphere software (Olea Medical Solution, La Ciotat, France). Volumetric T2 signal intensity of the entire lesion was evaluated at each time and the percentage of volume reduction (DV) point was calculated by the following equations:
DVmid : Vpre  Vmid =Vpre  100

Material and methods




DVpost : Vpre  Vpost =Vpre  100 where Vpre, Vmid and Vpost were the cancer volumes at pre, mid and postMRI, respectively. At postMRI, a radiologist with 8 years’ experience evaluated the degree of response of rectal cancer to CRT according to MRI-based TRG (mrTRG).15

Population Neoadjuvant CRT Patients (1) with biopsy-proven primary rectal adenocarcinoma and (2) locally-advanced disease at combined endorectal ultrasound and MRI evaluation (T3, and/or positive nodal status), and (3) who received neoadjuvant CRT followed by surgical resection were enrolled prospectively. Patients (1) with mucinous adenocarcinoma, (2) contraindications to MRI, and (3) who did not undergo surgery at San Raffaele Hospital (Milan, Italy) were excluded. The study was approved by the Institutional Review Board and written informed consent was obtained from all patients.

Chemotherapy was administered as previously described by Passoni et al.3 All patients received 100 mg/m2 oxaliplatin at days e14, 0, and þ14 associated with 200 mg/m2/ day 5-fluorouracil (5-FU) from day e14 to the end of chemotherapy (radiation therapy started on day 0). The planning target volume (PTV) of the CRT, including the tumour, mesorectum, and regional lymph nodes, was defined at preMRI and an additional “adaptive” PTV was generated by adding a 5-mm margin to the residual tumour identified at midMRI.

A. Palmisano et al. / Clinical Radiology 73 (2018) 555e563

Radiation therapy was administered daily and consisted of 18 fractions (2.3 Gy per fraction; total dose 41.4 Gy). In the last six fractions, an additional boost of 2.9e3.1 Gy per fraction was delivered to the adaptive PTV.

Histopathological evaluation The standard of reference for the assessment of histopathological response to CRT were tumour staging according to UICC TNM 7th edition and tumour downstaging by TRG € del classification16: TRG 0 when fibrosis was according to Ro completely absent, TRG 1 in case of fibrosis <25% of tumour mass, TRG 2 when tumour regression was 25e50%, TRG 3 when tumour regression >50% with fibrosis outgrowing the tumour mass, and finally, TRG 4 when no viable tumour cells were detected. TRG 4 represented complete responders (CR), patients with TRG 3 were considered partial responders (PR), whereas all the other grades (TRG 0e2) were nonresponders (NR).

Statistical analysis Numerical variables were described as means and standard deviations and graphically represented by boxplots. Categorical variables were summarised by frequencies and percentages. Differences between means were assessed by Welch’s t-test or analysis of variance (ANOVA) test. The overall performances of Vpre, Vmid, Vpost, DVmid, and DVpost were evaluated separately by receiver operating characteristic (ROC) curve and associated area under the curve (AUC). Firstly, the diagnostic performance of each variable in the prediction of CR versus PRþNR and NR versus PRþCR was investigated. The optimal cut-off maximising the association, measured by the associated pvalues, with outcome, was identified fitting a conditional inference tree model. Then, for each variable, the associated sensitivity, specificity, positive predictive value, negative predictive value, and accuracy evaluated. A multivariate model was then computed. Secondly, CR versus PR versus NR fitting a unique multivariable conditional inference tree model was considered. A recursive partitioning based on conditional permutation tests was applied: at each step the variable with the strongest significant association to the outcome is added and p-values were adjusted for multiplicity according to Benjamini and Yekutiely. This statistical approach prevented overfitting and overgrown trees, and no further pruning was needed.17,18 p-Values were computed by permutation methods to avoid any distributional assumption or asymptotic approximation, then p-values were adjusted for false discover rate and p<0.05 was considered significant.

Results Patients, treatment, and histopathological findings Forty-eight consecutive patients (female: male¼19: 29, mean age: 61 years; range 33e79 years) with biopsy-

557

proven LARC who were considered candidates for neoadjuvant CRT were enrolled (Table 1). Clinical staging, assessed by a combined MRI and endoscopic ultrasound evaluation, was mainly T3Nþ (30/48 patients T3Nþ; 12/48 patients T4Nþ; 2/48 patients T2Nþ; 4/48 patients T3N0). The tumours were located in the distal rectum, within 0e5 cm from the anal verge, in 19 patients (39.6%), in the middle rectum, within 5e10 cm from the anal verge, in 20 patients (41.7%), and in the proximal rectum, above 10 cm from the anal verge, in nine patients (18.7%). All patients underwent surgical resection after CRT (average: 11 weeks; range: 7e19 weeks) consisting mostly in low anterior resection (35/48 patients, 72.9%; Table 1). Pathological downstaging occurred in 30 patients (62.5%) with complete regression of rectal cancer in 13/48 (27.1%) with ypT0N0. The postoperative pathological stage was ypStage 0 in 13 patients (27.1%), ypStage I in 11 (22.9%), ypStage II in 14 (29.1%), and ypStage III in 10 (20.8%; Table 1). According to ypTRG, 13/48 (27%) patients were classified as CR to CRT, 25/48 (52%) were classified as PR, and the remaining 10 patients (22%) were NR.

MRI findings At pretreatment MRI, tumour volume in PR (50
47 cm3) was slightly higher than in NR (37
28 cm3), and CR (25
15 cm3), but these differences were not statistically significant. Table 1 Population features. Gender, n (%) Male Female Age, median (range, years) Tumour location, n (%) Distal rectum Middle rectum Proximal rectum Operative procedure, n (%) Low anterior resection Ultra-low anterior resection Intersphinteric resection Abdominoperineal resection Clinical T classification, n (%) 2 3 4 Clinical N classification, n (%) N0 Nþ ypT classification, n (%) 0 1 2 3 4 ypN classification, n (%) 0 1 or 2 ypTNM stage, n (%) 0 I II III

29 (60.4) 19 (39.6) 61 (33e79) 19 (39.6) 20 (41.7) 9 (18.7) 35 (72.9) 6 (12.5) 2 (4.2) 5 (10.4) 2 (4.2) 34 (70.8) 12 (25) 4 (8.3) 44 (91.7) 13 (27.1) 4 (8.3) 10 (20.8) 19 (39.6) 2 (4.2) 38 (79.2) 10 (28.8) 13 11 14 10

(27.1) (22.9) (29.1) (20.8)

558

A. Palmisano et al. / Clinical Radiology 73 (2018) 555e563

The volume of the tumour varied among the three groups both at midMRI (CR: 5.5
6.1 versus PR: 19.7
21.5 versus NR: 27.2
26.9, p¼0.0296; Fig 1a) and postMRI (CR: 3.3
4.4 versus PR: 7.9
7.1 versus NR: 15.5
12.2, p¼0.00242; Fig 1b). DV was significantly different among the three classes of response at midMRI (DVmid CR: 80
10% versus PR: 56
19% versus NR: 28
22%, p¼2.21016) as well as at postMRI (DVpost CR: 89
7% versus PR: 81
11% versus NR: 53
24%, p¼2.21016; Fig 1c). At preMRI, volumetric T2 signal intensity was slightly higher in CR (589
187) and PR (594
183) than NR (508
179), without reaching statistical significance (p¼0.7035). Similar T2 signal intensities were found at midMRI (CR: 544
188, PR: 584
195, NR: 534
187, p¼0.7989), whereas at postMRI, CR showed slightly, but not significantly, lower values than PR and NR (CR 395
191, PR 498
188, and NR 509
188, p¼0.7035). Fig. 2 shows the modification of volumetric T2 signal intensity during imaging monitoring in each group. T2 signal intensities showed a progressive minimal reduction in the CR and PR over time, without reaching statistical significance (p¼0.6066 and p¼0.1656, respectively). NR had similar T2 signal intensities at each time point (p¼0.7296). According to mrTRG, 7/48 patients were classified as CR (15%), 37/48 as PR (77%) and 4/48 as NR (8%). A substantial agreement between mrTRG and ypTRG was found (Cohen’s k: 0.628).

Early prediction of response to CRT Univariate analysis results regarding the role of tumour volume assessed by MRI in the prediction of CR and NR are reported, respectively, in Tables 2 and 3, in terms of optimal cut-off for each variable and relative sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy. Vmid, Vpost, and DVmid were selected as the most accurate predictors of CR; and DVmid and DVpost the most accurate predictors of NR. At multivariate analysis, Vmid and DVmid were selected as the best predictors to distinguish CR patients from the remaining patient population (PR þ NR). In particular, the combination of Vmid and DVmid enables a prediction of being a CR to be made with an overall accuracy of 92% (specificity 100%, PPV 100%, sensitivity 69%, NPV 90%; Fig 3a). Similarly, the combined evaluation of DVmid and Vpost was selected by the multivariate analysis as the best predictor of being a NR versus PR þ CR; its combination (Fig 3b) reaches an overall accuracy of 88% (specificity 100%, PPV 100%, sensitivity 50%, NPV 88%). In order to simultaneously distinguish CR versus PR versus NR, multivariate analysis found showed that the best accuracy is obtained by a two-step flow-chart (overall accuracy of 81.5%) obtained by the combined evaluation of Vmid and DVmid: Vmid <4.8 cm3 identifies CR; in case of Vmid >4.8 cm3, DVmid must be calculated, and patients are subsequently classified as NR or PR when the DVmid is <22% or >22%, respectively (Fig 4). According to the above-

mentioned flow-chart, midMRI correctly predicted 10/13 CR, 24/25 PR, and 5/10 NR (Fig 4). Vpre did not correlate with TRG; on the contrary, Vmid and Vpost were inversely correlated with TRG (r for Vmid¼e0.561 and for Vpost¼e0.617, p<0.001). A correlation between TRG and volume modification over time was also found. DVmid correlated with TRG (r¼0.713, p<0.001), which was slightly better than DVpost (r¼0.645, p<0.001).

Discussion MRI has high accuracy in the staging of rectal cancer because of its high soft-tissue resolution for the depiction of the primary tumour, lymph nodes, and mesorectal fascia8; however, the accuracy of conventional MRI in LARC restaging after CRT is reduced19 because of the difficulty in distinguishing fibrotic and inflammatory tissue from residual tumours. In order to increase the chance of recovery and improve oncological outcome, it is important to determine beforehand what kind of response to CRT patients will achieve, in order to assign patients to the appropriate treatment strategy. Nowadays, most clinical studies focus on the potential role of tumour reduction rate at post-treatment MRI, and they demonstrate that a higher tumour volume reduction rate on MRI after neoadjuvant therapy is associated with a better pathological response20e22 as well as a more favourable outcome12; however, this approach may fail to distinguish CR from PR, that are frequently characterised by similar DVpost.11,13 Therefore, in order to effectively guide patient-tailored treatments, a reliable and early assessment of the treatment response is required. For the early evaluation of the response to neoadjuvant treatment in breast cancer, serial MRI monitoring with an additional MRI performed during treatment was proposed.23 In confirming the usefulness of these findings, Rigter et al.23 suggested adapting chemotherapy schemes in relation to the degree of initial response to treatment evaluated at MRI obtained during neoadjuvant therapy, providing a consequent improvement of chemotherapy efficacy on less responsive breast cancers. In the present study, the potential of a midMRI performed during neoadjuvant CRT in LARC for the early assessment of tumour response was investigated. In this study (a) pretreatment tumour volume was not associated with tumour regression at histopathology, (b) midMRI seems to offer useful early information about tumour shrinkage, (c) the combined assessment of Vmid and DVmid appear effective in the stratification of response to CRT, being able to distinguish CR, PR, and NR early and with good accuracy. It is known that larger tumour volume is related to resistance to CRT because of more extensive necrosis and hypoxia24; however, in agreement with previous experience,12 pretreatment LARC volume was not effective in predicting postoperative pathological CRT response. Conversely, Vmid and Vpost were both associated with ypTRG. The present results suggest that tumour shrinkage mainly occurred shortly after the beginning of CRT, hence

A. Palmisano et al. / Clinical Radiology 73 (2018) 555e563

559

Figure 1 High-resolution T2-weighted images in one exemplifying patient for each response to CRT obtained before (preMRI), during (midMRI), and after (postMRI) treatment. These images clearly represent the different amounts of tumour shrinkage observed in relation to pathological complete response, partial response, and non-response and were characterised by different tumour volumes at midMRI (a) as well as at postMRI (b), and had different trends of volume reduction over time (c).

the assessment of tumour volume changes at midMRI may offer useful information about the different kinetics of tumour regression. CR showed an almost complete disappearance of viable cancer at midMRI while PR and NR showed a less significant tumour volume reduction. At

postMRI, both PR and CR showed a DVpost >80%, in agreement with previous literature11; moreover, the low specificity (63%) and PPV (48%) in the prediction of CR reflect the known challenge concerning the identification of viable residual cancer at postMRI. In fact, the presence of postCRT

560

A. Palmisano et al. / Clinical Radiology 73 (2018) 555e563

Figure 2 T2 signal intensity modification over time in CR (a), PR (b), and NR (c) Tumour T2 signal intensity was not significantly different according to pathological response at each time point. CR and PR showed a minimal progressive reduction of T2 signal intensity over time; however, these findings did not reach statistical significance.

Table 2 Univariate analysis results regarding magnetic resonance imaging (MRI) volumetric findings for the prediction of complete responders. MRI parameters

Cut-off Value

Sensitivity

Specificity

NPV

PPV

Accuracy

p-Value

Vpre Vmid Vpost DVmid DVpost

<21.55 cm3 <4.8 cm3 <2.9 cm3 >81% >83%

61.5% 77% 69% 62% 92%

80% 97% 89% 97% 63%

85% 91% 89% 87% 96%

53% 91% 69% 89% 48%

75% 92% 84% 87.5% 71%

0.831 0.044 <0.001 <0.001 <0.001

Vpre, tumour volume at pre-treatment MRI; Vmid, tumour volume at midMRI; Vpost, tumour volume at post-treatment MRI; Dmid, tumour volume reduction rate at midMRI; Dpost, tumour volume reduction rate at postMRI.

Table 3 Univariate analysis results about magnetic resonance imaging (MRI) volumetric findings for the prediction of non-responders. MRI parameters

Cut-off value

Sensitivity

Specificity

NPV

PPV

Accuracy

p-Value

Vpre Vmid Vpost DVmid DVpost

>26.3 cm3 >11.05 cm3 >6.8 cm3 <22% <65%

60% 90% 90% 50% 70%

63% 53% 74% 97% 97%

86% 95% 97% 88% 92.5%

30% 33% 47% 83% 87.5%

62.5% 60% 77% 87.5% 92%

0.021 <0.001 <0.001 <0.001 <0.001

Vpre, tumour volume at pre-treatment MRI; Vmid, tumour volume at midMRI; Vpost, tumour volume at post-treatment MRI; Dmid, tumour volume reduction rate at midMRI; Dpost, tumour volume reduction rate at postMRI.

inflammation and desmoplastic reaction extent to the perirectal fat interferes with the correct assessment of residual cancer, with a consequently large uncertainty in MRI volumetry.24,25 This result occurs in spite of good diagnostic confidence (Cohen’s kappa: 0.63 between mrTRG at postMRI and ypTRG), that is not significantly different from that reported in the MERCURY study (Cohen’s kappa: 0.65).15 The present study demonstrates the efficacy and usefulness of midMRI evaluation for the early prediction of LARC response to CRT and defines the best model to predict response based on MRI volumetric evaluation: tumours characterised by very small Vmid (<4.8 cm3) could be considered as having a complete response, while in cases of larger volume, DVmid should be evaluated, and patients

considered as PR or NR in cases of DVmid >22% or <22%, respectively. According to this flow-chart, patients at risk of treatment failure or incomplete response could be identified earlier and are potentially candidates for response-adapted radiotherapy strategies that increase the boost delivery in relation to the amount of persistent viable tumour to increase the chance of complete regression.26 Finally, this method may also enable the identification of those who will likely achieve a complete response, and could be candidates for a conservative strategy with serial MRI monitoring in the context of a “wait and see” policy.27 In a conservative “wait and see” approach, serial MRI monitoring may also enable the early identification of local relapse, as well as the

A. Palmisano et al. / Clinical Radiology 73 (2018) 555e563

561

Figure 3 Diagnostic flowchart for the early prediction of CR (a) and NR (b), separately. (a) Multivariate analysis for the prediction of a CR versus PRþNR suggested a two-step diagnostic flowchart, characterised by an overall accuracy of 92%, based on the evaluation of tumour volume at midMRI followed by the assessment of tumour volume reduction rate at midMRI. (b) For the prediction of a NR versus CR þ PR, a diagnostic flowchart including the combined evaluation of Dmid and Vpost, with an overall accuracy of 88% was proposed.

possible persistence of viable tumour, as occurred in a single patient of this population wrongly classified as a CR at midMRI. An additional midMRI may be useful for the clinical management of patients in order to avoid overtreatment and to improve treatment success with a more tailored CRT

approach. These findings will need to be confirmed in a larger prospective study and the potential additive role of a mid-term multiparametric MRI evaluation should be investigated. In conclusion, the implementation of MRI monitoring of rectal cancer response to CRT adding an intermediate MRI,

562

A. Palmisano et al. / Clinical Radiology 73 (2018) 555e563

Figure 4 Diagnostic flowchart based on multivariate results for the early stratification of patients in relation to the different degrees of response to treatment. The proposed method (overall accuracy 81.5%) allows CR versus PR versus NR to be distinguished early based of Vmid and Dmid.

could offer useful information about the early trend of response and predict ypTRG with high accuracy. It could be useful in tailoring treatment, with eventual adaptation of the CRT scheme in order to increase the chance of recovery. Obtaining crucial information for the management of patients based on the addition of a single MRI during treatment could indeed be highly beneficial, especially considering the low cost of the examination (no contrast media is required) and the short acquisition time.

References 1. Salem ME, Hartley M, Unger K, et al. Neoadjuvant combined-modality therapy for locally advanced rectal cancer and its future direction. Oncology (Williston Park) 2016;30(6):546e62. 2. Brown CL, Ternent CA, Thorson AG, et al. Response to preoperative chemoradiation in stage II and III rectal cancer. Dis Colon Rectum 2003;46(9):1189e93. 3. Passoni P, Fiorino C, Slim N, et al. Feasibility of an adaptive strategy in preoperative radiochemotherapy for rectal cancer with image-guided tomotherapy: boosting the dose to the shrinking tumor. Int J Radiat Oncol Biol Phys 2013;87(1):67e72. 4. Habr-Gama A, Perez RO, Nadalin W, et al. Operative versus nonoperative treatment for stage 0 distal rectal cancer following chemoradiation therapy: long-term results. Ann Surg 2004;240(4):711e7. 5. Vaccaro CA, Yazyi FJ, Ojra Quintana G, et al. Locally advanced rectal cancer: preliminary results of rectal preservation after neoadjuvant chemoradiotherapy. Cir Esp 2016;94(5):274e9. 6. Onaitis MW, Noone RB, Hartwig M, et al. Neoadjuvant chemoradiation for rectal cancer: analysis of clinical outcomes from a 13-year institutional experience. Ann Surg 2001;233(6):778e85. 7. MERCURY Study Group. Extramural depth of tumor invasion at thinsection MR in patients with rectal cancer: results of the MERCURY study. Radiology 2007;243(1):132e9.

8. Tudyka V, Blomqvist L, Beets-Tan RG, et al. EURECCA consensus conference highlights about colon and rectal cancer multidisciplinary management: the radiology experts review. Eur J Surg Oncol 2014;40(4):469e75. 9. Blazic I, Campbell NM, Gollub MJ. MRI for evaluation of treatment response in rectal cancer. Br J Radiol 2016;89(1064):20150964. 10. van der Paardt MP, Zagers MB, Beets-Tan RG, et al. Patients who undergo preoperative chemoradiotherapy for locally advanced rectal cancer restaged by using diagnostic MR imaging: a systematic review and meta-analysis. Radiology 2013;269(1):101e12. 11. Neri E, Guidi E, Pancrazi F, et al. MRI tumor volume reduction rate versus tumor regression grade in the pre-operative re-staging of locally advanced rectal cancer after chemo-radiotherapy. Eur J Radiol 2015;84(12):2438e43. 12. Yeo SG, Kim DY, Park JW, et al. Tumor volume reduction rate after preoperative chemoradiotherapy as a prognostic factor in locally advanced rectal cancer. Int J Radiat Oncol Biol Phys 2012;82(2):e193e9. 13. Kim YH, Kim DY, Kim TH, et al. Usefulness of magnetic resonance volumetric evaluation in predicting response to preoperative concurrent chemoradiotherapy in patients with resectable rectal cancer. Int J Radiat Oncol Biol Phys 2005;62(3):761e8. 14. Martens MH, van Heeswijk MM, van den Broek JJ, et al. Prospective, multicenter validation study of magnetic resonance volumetry for response assessment after preoperative chemoradiation in rectal cancer: can the results in the literature be reproduced? Int J Radiat Oncol Biol Phys 2015;93(5):1005e14. 15. Patel UB, Taylor F, Blomqvist L, et al. Magnetic resonance imagingdetected tumor response for locally advanced rectal cancer predicts survival outcomes: MERCURY experience. J Clin Oncol 2011;29(28):3753e60. € del C, Hofheinz R, Liersch T. Rectal cancer: state of the art in 2012. 16. Ro Curr Opin Oncol 2012;24:441e7. 17. Hothorn T, Hornik K, Zeileis A. Unbiased recursive partitioning: a conditional inference framework. J Comput Graph Stat 2006;15(3):651e74. 18. Miotto P, Cabibbe AM, Feuerriegel, et al. Mycobacterium tuberculosis pyrazinamide resistance determinants: a multicenter study. MBio 2014;5(5). e01819e14. 19. Hanly AM, Ryan EM, Rogers AC, et al. Multicenter Evaluation of Rectal cancer ReImaging pOst Neoadjuvant (MERRION) therapy. Ann Surg 2014;259(4):723e7.

A. Palmisano et al. / Clinical Radiology 73 (2018) 555e563 20. Kim NK, Baik SH, Min BS, et al. A comparative study of volumetric analysis, histopathologic downstaging, and tumor regression grade in evaluating tumor response in locally advanced rectal cancer following preoperative chemoradiation. Int J Radiat Oncol Biol Phys 2007;67(1):204e10. 21. Han YB, Oh SN, Choi MH, et al. Clinical impact of tumor volume reduction in rectal cancer following preoperative chemoradiation. Diagn Interv Imaging 2016 Sep;97(9):843e50. 22. Sathyakumar K, Chandramohan A, Masih D, et al. Best MRI predictors of complete response to neoadjuvant chemoradiation in locally advanced rectal cancer. Br J Radiol 2016;89(1060):20150328. 23. Rigter LS, Loo CE, Linn SC, et al. Neoadjuvant chemotherapy adaptation and serial MRI response monitoring in ER-positive HER2-negative breast cancer. Br J Cancer 2013;109(12):2965e72.

563

24. Kuo LJ, Chern MC, Tsou MH, et al. Interpretation of magnetic resonance imaging for locally advanced rectal carcinoma after preoperative chemoradiation therapy. Dis Colon Rectum 2005;48(1):23e8. 25. Kye BH, Kim HJ, Kim G, et al. Multimodal assessments are needed for restaging after neoadjunvant chemoradiation therapy in rectal cancer patients. Cancer Res Treat 2016;48(2):561e6. 26. Stuyck C, Wegge M, Bulens P, et al. Moderate dose escalation with volumetric modulated arc therapy improves outcome in rectal cancer. Acta Oncol 2017;56(11):1501e6. 27. Yang TJ, Goodman KA. Predicting complete response: is there a role for non-operative management of rectal cancer? J Gastrointest Oncol 2015;6(2):241e6.