Diffusion-weighted imaging complements T2-weighted MRI for tumour response assessment in squamous anal carcinoma

Institutional board waiver of informed consent was obtained for this retrospective study of consecutive MRI data obtained as part of the standard care pathway. Patients with biopsy-proven squamous cell carcinoma of the anal canal undergoing treatment with definitive chemoradiotherapy between February 2009 and May 2020 were identified from the Picture Archiving and Communication System (PACS) and electronic patient record (EPR) of a tertiary care institution (Guy’s and St Thomas’ NHS Foundation Trust). Inclusion criteria were baseline and post treatment MRI (within 3 months of treatment completion) available from PACS; TNM 8th ed. T2 stage; or greater tumours [15], equivalent to tumour diameter >2 cm. Exclusion criteria were absence of DWI on baseline or post treatment MRI; DWI of insufficient diagnostic quality; no visible tumour on baseline MRI; and prior tumour surgical excision. The patient flowchart is shown in Fig. 1.

Radiotherapy was delivered to a mean dose of 50.86 Gy (range 50.4–54 Gy) using a linear accelerator (Elekta or Varian) applying a 3D conformal or intensity-modulated technique. Concomitant chemotherapy consisted of mitomycin C (12 mg/m² on day 1) with either 5-fluorouracil (1000 mg/m²/day, continuous venous infusion, on days 1–4 and 29–32) or capecitabine orally (825 mg/m² twice a day on radiation days).

Following completion of chemoradiotherapy, patients were clinically assessed at 8–10 weeks, then every 3 months for the first 2 years, and every 6 months afterwards as per standard institutional practice. Endoscopic evaluation ± MRI/CT was undertaken if canal ± locoregional/distant recurrence was suspected clinically.

Patients were scanned supine on one of five 1.5- or 3.0-Tesla MRI scanners (Magnetom Avanto, Aera or Skyra, Siemens Healthineers) using an 18-channel pelvic phased array coil. The examination protocol included axial and sagittal T2-weighted turbo spin-echo (TSE) sequences covering the whole pelvis, and high-resolution small-field-of-view T2-weighted TSE sequences perpendicular and parallel to the anal canal. DWI consisted of a single-shot spin-echo echo planar imaging (EPI) axial diffusion-weighted sequence encompassing the pelvis with three b-values (0, 100, 800 s/mm²). Apparent diffusion coefficient (ADC_0-800) maps and calculated high b-value images (b = 1400 or 1600 s/mm²) were created automatically at the time of acquisition. Patients did not undergo any additional preparation prior to the examination. Typical acquisition parameters are summarised in Supplemental Table 1.

Baseline and post treatment MRI were evaluated sequentially by four independent observers, blinded to clinical outcome: two senior radiology residents (K.M., A.A.), with 1 year’s experience in oncologic pelvic MRI but no specific experience in staging squamous anal carcinoma (referred to as ‘non-expert observers’) and two subspecialist radiologists (D.P., V.G.) with over 10 years’ experience (‘expert observers’).

Anonymised scans were presented in a randomised order. Post treatment MRIs were assessed next to baseline MRIs on dual monitors (Sectra IDS7 workstations, Sectra AB). The first reading session was based on multiplanar T2-weighted sequences alone. Tumour response was evaluated according to a previously published 5-point tumour regression grade (TRG) score [10], outlined in Table 1. In addition to TRG, the following primary tumour characteristics were recorded: size (maximum diameter in any plane); location (lower canal, mid canal, upper canal/anorectum); invasion of adjacent structures (prostate, vagina, ischioanal fossa). The second reading session, separated from the first by a 12-week wash out period, included T2-weighted and DWI assessed in conjunction. Multiplanar T2 sequences, acquired/calculated b-value images and corresponding ADC maps were displayed simultaneously. DWI image quality was scored subjectively as inadequate, adequate or good by each observer, documenting the nature of image degradation as free text, when present. Tumour response was re-scored according to a modified DWI-TRG system, outlined in Table 1. Post treatment DWI images were regarded positive for residual disease when diffusion restriction (signal hyperintensity on high b-value images matched by hypointensity on ADC map) remained present at the site of the tumour, excluding linear diffusion restriction spatially matched to anorectal mucosa. A single-slice, free-hand region of interest was drawn around the tumour on ADC maps, with reference to the corresponding DWI and T2-weighted images, and the mean tumour ADC value recorded for baseline MRI. During both reading sessions, observers scored their subjective confidence in assessing tumour response on a scale of 1 (low) to 5 (high).

Statistical analyses were performed by a senior statistician (P.B.) using Stata (v15.1; StataCorp LP). Normally distributed variables were expressed as mean ± standard deviation. Categorical variables were expressed as absolute numbers and their percentages. The McNemar test was used to compare the number of indeterminate TRG vs. other TRG scores between reading sessions. Interobserver agreement was assessed using the kappa statistics (kappa < 0.21 = poor agreement; 0.21–0.40 = fair; 0.41–0.60 = moderate; 0.61–0.80 = good; > 0.80 = excellent). Kappa values and their standard errors were used to perform a z-test to compare the level of agreement between reading sessions. The Wilcoxon matched-pairs test was used to compare observer confidence scores. Analyses were performed for each observer separately, and for all observers combined. A p value <.05 was taken to represent statistical significance.

Baseline patient characteristics are summarised in Table 2. The final cohort consisted of 85 patients, ranging in age between 34 and 86 years (mean, 59 years ± 12 [SD]; 55 women). Mean tumour size was 5.1 ± 2.1 cm. A large proportion of patients had locally advanced disease at baseline (53%, 45/85), defined as T3 stage or greater, and/or tumours located in the upper canal/anorectum (52%, 44/85). Mean clinical follow-up duration was 32 ± 18 months. Clinical disease recurrence was recorded in 36% (31/85) of patients: local, in 18% (15); nodal, in 9% (8); metastatic, in 14% (12). Clinical complete tumour response was recorded in 91% (77/85) of patients at 8–10 weeks from the end of treatment. There was partial response in 6% (5/85). Progressive disease was recorded in 4% (3/85). Local recurrence was documented in 8% (7/85) of patients: 5% (4), corresponding to late recurrence, beyond 12 months from the end of treatment; 4% (3), corresponding to microscopic subclinical recurrence at 6–9 months from the end of treatment. Four patients underwent salvage surgery by means of abdominoperineal excision of the rectum.

DWI image quality was scored as inadequate in 3 cases, which were excluded; adequate in 26% (22/85); and good in 74% (63/85). The most common problems affecting DWI quality were low signal-to-noise ratio and susceptibility artefacts, particularly from air/gas at the anal margin and in the rectal lumen. Calculated b-value (b1400 or b1600 s/mm²) images were available in 70/85 (82%) cases. Baseline tumour conspicuity on b800 images and ADC maps was high in all included cases, with mean tumour ADC values of 0.910 ± 0.182 × 10⁻³ mm²/s.

TRG scores from the four radiologists are shown in Fig. 2. With the inclusion of DWI, the number of indeterminate TRG-3 scores decreased significantly for three of the four radiologists examined individually (difference range, 11–19%; p range, < 0.001–0.04), and for all radiologists combined (difference, 12%; p < 0.001) (Table 3). For all observers combined, the number of TRG-3 cases halved from 24% (82/340) of the total based on T2-weighted MRI alone to 12% (41/340) based on T2-weighted plus DWI. Indeterminate TRG-3 scores changed most frequently to DWI-TRG-2 (41%, 34/82), corresponding to excellent response (Figs. 3 and 4); 9% (7/82) changed to DWI-TRG-4, corresponding to minimal response (Fig. 5). The remaining 50% of TRG-3 (41/82) corresponded to indeterminate DWI-TRG-3 scores (Fig. 6).

Observers’ confidence in assessing response increased with the addition of DWI. Scores were higher for each of the four observers (p < 0.001), and for all observers combined (Supplemental Table 2). For all observers combined, 84% (287/340) of confidence scores were 4 or 5 for T2-weighted plus DWI, compared to 55% (188/340) for T2-weighted MRI alone.

Interobserver agreement was between fair and moderate (Supplemental Table 3), with kappa values ranging between 0.28 and 0.58. The highest agreement was achieved by non-expert observers assessing response on T2-weighted plus DWI. No significant differences in interobserver agreement were found between the two response assessment methods (p = 0.16–0.40).

Correspondence between MRI TRG scores and clinical tumour response is summarised in Table 4. Patients with complete clinical response (n = 77) were assigned TRG scores of 1 or 2 in 88% (270/308) of cases using T2-weighted plus DWI, versus 73% (226/308) based on T2-weighted MRI alone. Patients with partial clinical response (n = 5) were assigned scores of 3 or above in 75% (15/20) and 65% (13/20) of cases respectively. Eleven out of 12 TRG scores in 3 patients with progressive disease were ‘4’ or ‘5’. In 7 patients with documented local recurrence after an initial complete response, 64% (18/28) TRG-2 and 93% (26/28) DWI-TRG-2 scores were recorded, in line with complete responders without subsequent recurrence.