Multi-Institutional Audit of FLASH and Conventional Dosimetry With a 3D Printed Anatomically Realistic Mouse Phantom

Abstract

Purpose: We conducted a multi-institutional dosimetric audit between FLASH and conventional dose rate (CONV) electron irradiations by using an anatomically realistic 3-dimensional (3D) printed mouse phantom.

Methods and materials: A computed tomography (CT) scan of a live mouse was used to create a 3D model of bony anatomy, lungs, and soft tissue. A dual-nozzle 3D printer was used to print the mouse phantom using acrylonitrile butadiene styrene (∼1.02 g/cm³) and polylactic acid (∼1.24 g/cm³) simultaneously to simulate soft tissue and bone densities, respectively. The lungs were printed separately using lightweight polylactic acid (∼0.64 g/cm³). Hounsfield units (HU), densities, and print-to-print stability of the phantoms were assessed. Three institutions were each provided a phantom and each institution performed 2 replicates of irradiations at selected anatomic regions. The average dose difference between FLASH and CONV dose distributions and deviation from the prescribed dose were measured with radiochromic film.

Results: Compared with the reference CT scan, CT scans of the phantom demonstrated mass density differences of 0.10 g/cm³ for bone, 0.12 g/cm³ for lung, and 0.03 g/cm³ for soft tissue regions. Differences in HU between phantoms were <10 HU for soft tissue and bone, with lung showing the most variation (54 HU), but with minimal effect on dose distribution (<0.5%). Mean differences between FLASH and CONV decreased from the first to the second replicate (4.3%-1.2%), and differences from the prescribed dose decreased for both CONV (3.6%-2.5%) and FLASH (6.4%-2.7%). Total dose accuracy suggests consistent pulse dose and pulse number, although these were not specifically assessed. Positioning variability was observed, likely due to the absence of robust positioning aids or image guidance.

Conclusions: This study marks the first dosimetric audit for FLASH using a nonhomogeneous phantom, challenging conventional calibration practices reliant on homogeneous phantoms. The comparison protocol offers a framework for credentialing multi-institutional studies in FLASH preclinical research to enhance reproducibility of biologic findings.

FIGURE 1.

Outline of the general workflow to develop the 3D printed mouse phantom. Bone, lungs, and soft tissue were contoured by using a 3D micro-CT dataset. The contours were converted to mesh files, imported into CAD software, and edited by splitting the phantom in half and including registration pegs for radiochromic film positioning. Printing parameters were optimized, and material densities were verified by weighing cubes of known volume. Mesh files were printed by using dual simultaneous extrusion with polylactic acid (PLA) and acrylonitrile butadiene styrene (ABS) printing materials to mimic bone and soft tissue, respectively. Lightweight PLA was used to mimic lungs. Radiochromic films were laser-cut to cover different anatomic sites and ensure accurate positioning within the phantom.

FIGURE 2.

Irradiation geometries for the three institutions. In panel (a), a clinical linear accelerator was configured at Stanford to deliver FLASH dose rates. The gantry cover was removed to allow the cradle to be placed near the scattering foil. The beam enters from underneath the phantom. The mouse phantom was placed in organ-specific 3D printed collimators housed inside a cradle. The Body Long and the Body Short collimators are shown here, with the red boxes indicating the irradiated area. (b) The irradiation geometry at MD Anderson. A FLASH-enabled interoperative electron linac was used. 3D-printed organ-specific collimators were attached to the gantry head. The beam enters from the top. The whole abdomen collimator is shown in the middle panel and the whole lung collimator in the left panel. (c) The irradiation setup for CHUV shows setup for Body Long and Body Short irradiations.

FIGURE 3.

Film dosimetry calibrations. (a) Film calibration curve and 95% prediction interval. A 5^th order polynomial was used for fitting. (b) Changes in the dose response of netOD over time. (c) Percentage difference in net OD for the same dose from one time point to the next. The uncertainty associated with using a calibration curve across these different time points can be as high as ~2%. (d) Validation of the calibration curve by delivering known doses under reference conditions. The largest error between measured and delivered dose was 0.32 Gy for a prescribed dose of 23 Gy (1.4%). The green dashed line represents theoretical agreement for reference (slope = 1).

FIGURE 4.

Characterization of 3D mouse phantom. (a) 2D coronal slices of the CT scanned images of the mouse phantom, with region of interests (ROIs) (in red) used to compare the HU values between the reference CT of the live mouse and the 3D printed phantom. (b) Comparisons of mean HU values show small, calculated mass density differences of 0.1 g/cm³ for bone, 0.12 g/cm³ for lung, and 0.03 g/cm³ for soft tissue regions between the phantom and the live mouse. (c) 2D coronal slices of CT scans of two 3D printed phantoms. The three marked ROIs were used to quantify the difference in soft tissue (ROI1), lung (ROI2), and bone (ROI3) regions from one print to another. CT images for the 3^rd and 4^th phantom are shown in the Supplementary figures. The average differences in absolute HU across 4 phantoms (with phantom 1 as reference) were 8.8 HU for soft tissue, 59.7 HU for lung, and 10.7 HU for bone. (d) Line profiles for the 4 phantoms across the pixels indicated in green in (c), with differences in the profiles of the lines also shown. The average HU difference between the line profiles in (d) and (e) were 17.4 HU (d) and 21.9 HU (e) across the 4 phantoms (with phantom 1 as the reference), consistent with excellent print-to-print reproducibility. (e) Comparison of whole-lung irradiation between phantoms 1 and 2 indicates that small differences in HU values have minimal effects on dose distributions.

FIGURE 5.

2D dose comparison of FLASH vs CONV irradiations for the (A) Body Long, (B) Body Short, and (C) Brain fields for the three institutions (anonymized). Each institution performed two replicates, with 5 films irradiated per anatomic region in each replicate. For each field, the same ROIs were chosen on each film (red rectangles) to calculate the average dose difference. White lines indicate the regions that were used for line profiles. Both vertical and horizontal line profiles are shown for all irradiations; graphs on the left show vertical line profiles, and graphs on the right show horizontal line profiles. The difference between the CONV and FLASH profiles is shown under each graph. Error bars represent the standard error obtained from 5 repeated measurements. Summary statistics are shown in Table 2 and 3. Overall, discrepancies between FLASH and CONV, and discrepancies from the prescribed dose, were small across all institutions and tended to decrease between the first and second replicates.

FIGURE 5.

2D dose comparison of FLASH vs CONV irradiations for the (A) Body Long, (B) Body Short, and (C) Brain fields for the three institutions (anonymized). Each institution performed two replicates, with 5 films irradiated per anatomic region in each replicate. For each field, the same ROIs were chosen on each film (red rectangles) to calculate the average dose difference. White lines indicate the regions that were used for line profiles. Both vertical and horizontal line profiles are shown for all irradiations; graphs on the left show vertical line profiles, and graphs on the right show horizontal line profiles. The difference between the CONV and FLASH profiles is shown under each graph. Error bars represent the standard error obtained from 5 repeated measurements. Summary statistics are shown in Table 2 and 3. Overall, discrepancies between FLASH and CONV, and discrepancies from the prescribed dose, were small across all institutions and tended to decrease between the first and second replicates.