Ten patients were enrolled in the study, from 2020 to 2023, of which seven completed the entire course of the study. Post-BT images were not acquired for three patients and, as a result, are not reported herein. All participants provided written informed consent to a study protocol approved by a local Research Ethics Board. All participants with gynecologic cancer underwent whole pelvis radiation therapy (WPRT) and HDR BT. The inclusion criteria included: any patient eligible for BT internal implantation without MR guidance, life expectancy of greater than 6 months, and ECOG performance status of < 2.

All patients received WPRT of 45 Gy in 25 fractions (i.e. 1.8 Gy/fraction) using VMAT on an Elekta VersaHD LINAC (Elekta AB, Stockholm, Sweden). The HDR brachytherapy procedure was performed with the intent to comply to the American Brachytherapy Society consensus guidelines, with the most common physical dose prescription being 27.5 Gy in 5 fractions [17, 18]. The Venezia applicator (Elekta AB, Stockholm, Sweden) or the Syed template were used for treatment. In this cohort, patients were treated in the inpatient setting with either one procedure or two separate procedures one week apart. The treatment characteristics for each patient, including CTV D90Gy and the number of insertions are reported in Table 1.

MR images were acquired at three time points baseline/pre-RT, post-WPRT/pre-BT (one day after delivery of the prescribed WPRT-dose), and 3 or 6 months post-BT on 1.5T MRI scanners (either MAGNETOM Aera or MAGNETOM Sola, Siemens Healthineers GmbH, Erlangen, Germany) using torso and spine phased-array coils. No patient was scanned on multiple MRI systems.

IR-UTE MRI was performed using a stack-of-spirals dual-echo research application sequence [19]. Whole pelvis IR-UTE images were acquired in the coronal plane (parameters: echo times [TE1, TE2] /repetition time (TR)/flip-angle(Ɵ) = 0.05 ms, 2.69 ms/7 ms /8º, field-of-view [FOV] = 36–39 cm, time-duration/spiral = 1800µs, 220 spirals/image, signal-averages [AV] = 1, resolution = 0.9 × 0.9 × 2.5 mm3 or 1.0 × 1.0 × 2.5 mm³, 88 slices/scan with 50% sice under-sampling, scan-time [TA] = 246s).

Non-contrast IR-UTE images were acquired to evaluate FDiffuse with TInversion = 60 ms, with the inversion pulse used to null the fat signal and accentuate short T1 relaxation-time tissue components, while the LGE-IR-UTE images were acquired to evaluate FDense with TInversion = 200 ms, primarily to suppress vascular signal, approximately 15 min following contrast administration. All participants received a 0.1 mmol/kg dose of Gadavist (Gadobutrol, Bayer Radiology) via a power-injector. Herein we refer to the non-contrast IR-UTE images as FDiffuse images and the LGE-IR-UTE images as FDense images.

The residual pockets, herein referred to as the remnant tumor, were detected reliably with multi-parametric MRI (mpMRI) as previously shown [20, 21]. To detect the remnant living tumor(s) volume at the time-points before (pre-RT), after WPRT (pre-BT), and 3–6 months post-BT, 2D T2 weighted turbo spin-echo (TSE) images were acquired along axial and sagittal planes (parameters: TE/TR/Ɵ = 122 ms/3760 ms/120º, resolution = 1 × 1 × 2.5 mm3, bandwidth [BW] = 300 Hz/pixel, echo train length [ETL] = 14, AV = 2, TA = 379 s). Axial diffusion-weighted images (DWI) were acquired using a single-shot fat-suppressed SE-EPI sequence with two sagittal saturation slabs placed at the left and right 27 cm-FOV borders along the phase-encoding direction to prevent folding and reduce geometric distortion (parameters: monopolar diffusion-encoding, diffusion-directions = 3, b-values = 100, 500, 900 s/mm2, TE/TR/Ɵ = 92 ms/6000 ms/90º, ETL = 158, resolution 1.7 × 1.7 × 3.7 mm3, BW = 106 Hz/pixel, TA = 596s). Additionally, Dynamic Contrast Enhanced (DCE) fat-suppressed 3DGRE images were acquired continuously from 1 min before contrast injection to 5 min after contrast injection (parameters: TE/TR/Ɵ = 1.97 ms/4.5 ms/20º, resolution = 1.3 × 1.3 × 2.0mm3, BW = 400 Hz/pixel, accelerated quick fat-sat, centric-encode, average = 1, 60 slices/slab, TA = 20 s/dynamic-frame).

Remnant tumor segmentation was performed on the mpMRI data images by a radiologist (BD), with over 25 years abdomino-pelvic cancer experience. The baseline/pre-RT tumor was contoured on the baseline/pre-RT MR images. The remnant tumor present post-WPRT/pre-BT was contoured on the post-WPRT/pre-BT and post-BT images. To reduce the uncertainty introduced from image registration, the remnant tumor present post-BT was contoured on the post-BT images only. The high risk clinical target volume (HR-CTV) was contoured by an experienced radiation oncologist. T2-weighted (T2w) images acquired at the time of BT (which were used for BT treatment planning) were registered to the FDiffuse and FDense images acquired without the applicator post-BT (Fig. 1, Step 1). RayStation 2023B (RaySearch, Stockholm, Sweden) was used for biomechanical deformable image registration [22]. Briefly, the deformation is driven by controlling structures (i.e. uterus and cervix) that are defined in both the moving and the stationary image sets. Linear, elastic material properties are assigned based on structural information [23]. For patients receiving more than a single applicator insertion, each dose distribution was converted to equivalent dose in 2 Gy fractions (EQD2), and a composite dose volume was created in Raystation. As outlined in the ABS consensus guidelines on HDR BT for locally advanced cervical cancer [17], we utilized an alpha/beta ratio of 10 Gy for the target and 3 Gy for the OARs. An alpha/beta ratio of 10 Gy was used for fibrosis that formed in the target and an alpha/beta ratio of 3 Gy was used for fibrosis that formed elsewhere. The images acquired during the second insertion were deformably registered to the images acquired during the first insertion. This deformation field was applied to the dose and a sum dose volume was created.

Flow charts outlining the study workflow. Step 1 demonstrates image registration and the contour propagation workflow. Step 2 shows the creation of the responsive ROI and isodose line (IDL) rings

Contours corresponding to the 50%, 100%, 150%, and 200% isodose lines (IDLs) of the prescription BT-dose were generated on the T2-weighted images acquired at the time of BT. IDLs were used and scaled relative to prescription to allow for consistency across patients that had varying BT prescriptions. As shown in Fig. 1, Step 2b, the responsive irradiated region, the region where remnant tumor was not detected post BT, was determined by subtracting the post-BT remnant tumor from the post-WPRT/pre-BT HR-CTV. IDL rings were generated to encompass a dose range as follows: 50% Ring = 50–100% of prescription; 100% Ring = 100–150% of prescription, 150% Ring = 150–200% of prescription, 200% Ring > 200% of prescription. We also generated dose rings to encompass the following four dose ranges: 10–25 Gy, 25–40 Gy, 40–55 Gy, and over 55 Gy. The “responsive ROI IDLs” were defined as the overlap between the responsive ROI and the IDL/dose rings. This was done for all four IDL rings (Fig. 1, Step 2c).

The non-contrast and LGE-IR-UTE images were normalized based on the mean SI of a 1 cm diameter region-of-interest (ROI) in the gluteal-muscle (i.e. normalized as a percent of gluteal muscle SI), which is outside the prescribed irradiated region as well as a tissue with uniform fibrosis. We define the mean FDiffuse SI and FDense SI as the mean normalized SI derived from the non-contrast and LGE-IR-UTE images, respectively. Mean FDiffuse and FDense SI were computed for the baseline/pre-RT, post-WPRT/pre-BT remnant tumor, post-BT remnant tumor, and responsive regions. The mean FDense SI within the IDL rings were also determined. We further computed the mean FDense SI within the “responsive ROI IDLs”. Table 2 provides a summary of ROIs utilized for analysis.

Univariate relationships were evaluated using linear regressions (r2) and Pearson correlations (r), specifically to evaluate the relationship between dose delivered and the FDense SI of the irradiated remnant tumor and the FDiffuse SI of the 100% Ring. Paired t-test with Bonferroni correction was performed to evaluate the difference between SI within the IDL contours. All statistical tests were performed using SPSS 28.0 (IBM, Armonk NY). Results were considered significant when the probability of making a Type I error was less than 5% (P < .05).