The impact of bladder and rectal dynamics on prostate and seminal vesicles intrafraction motion and deformation in radiotherapy

Graphical abstract

Highlights

• •Magnetic Resonance Imaging features predict prostate and seminal vesicles motion.
• •Low baseline bladder volume (<190 mL) predicts greater intrafraction motion.
• •Rectum volume showed no strong correlation to prostate/seminal vesicles motion.
• •Fuller starting bladder (>332 mL) may allow for smaller margin for seminal vesicles.

Abstract

Introduction

Treatment uncertainties influenced by organ intrafraction motion complicate the widespread adoption of hypofractionated radiotherapy. This study aims to identify imaging features on pre-treatment magnetic resonance imaging (MRI) scans that describe prostate and seminal vesicle (SV) intrafraction motion, with the goal of informing and improving treatment planning.

Materials/methods

Thirty prostate cancer participants treated on an Elekta Unity 1.5T MR-Linac were recruited, with a series of volumetric MR images acquired pre-, during and post- treatment over multiple fractions. nnU-Net was used to automatically contour the prostate, rectum, SV and bladder. These contours quantified prostate and SV intrafraction motion and enabled extraction of imaging features. A linear regression model assessed relationships between the organs intrafraction motion, treatment margins, and the extracted features.

Results

Bladder filling during treatment influenced both SV and prostate intrafraction motion, especially, when baseline bladder volume was <190 mL for both prostate (R² = 0.142) and SV (R² = 0.258). Rectum volume showed no strong correlation with motion. Baseline bladder volume below 332 mL increased the required SV treatment margins to 5.8 mm, compared to 3.5 mm for larger volumes.

Conclusion

This study demonstrated that the baseline bladder volume at start of a treatment fraction predicts for both SV and prostate intrafraction motion, by mediating the effect of bladder filling, and that SV treatment margins could be reduced for a favourably sized bladder. These findings may support refining treatment protocols such as aiming for an initial bladder volume of at least 190 mL.

1. Introduction

Conventionally fractionated radiation therapy is an effective treatment for prostate cancer but organ intrafraction motion introduces uncertainty in the treatment area, potentially compromising treatment efficacy. To mitigate this, a safety margin is added to the clinical target to form the planning target volume (PTV) [1]. While a larger margin reduces the likelihood of a target undercoverage, it also exposes a larger volume of healthy tissue to radiation, thereby raising the risk of toxicity [2]. This concern is heightened in hypofractionated or ultra-hypofractionated (UH) radiation therapy treatments [3], [4], [5]. Where higher per-fraction doses amplify the risk of toxicity to organs-at-risk (OAR). Therefore, accurately understanding the extent and magnitude of organ intrafraction motion is critical for optimising PTV margins while maintaining treatment effectiveness.

Prostate intrafraction motion ranging from 3 mm to over 10 mm has been reported in the literature [6], [7] with motion tending to increase alongside treatment duration [7], [8], [9]. This behaviour is potentially problematic in the context of an UH regime, due to their extended treatment time, which may necessitate a larger treatment margin with potential for increased toxicity due to the high doses employed per fraction [2]. Treatment margins can be personalised and potentially reduced by incorporating predictors of prostate intrafraction motion, including body mass index [10], rectal diameter [11], type of rectal filling [12], bladder volume size [13], existence of endorectal balloons or hydrogel [14], respiratory motion [15] and body positioning [16].

Seminal vesicles (SV) are often incorporated into a prostate cancer treatment plan, yet research on SV intrafraction motion is relatively sparse, representing a significant research gap [17]. This often leads to additional expansion of the gross treatment volume to account for SV motion [18], [19], [20]. Previous studies had investigated factors such as the influence of bladder filling and rectal distension on SV and prostate motion [21], the correlation between prostate and SV motion [22], the effect of tumour invasion into SV [23], the influence of hydrogel spacer [24], [25], [26] and the influence of an endorectal balloon [14], [27].

In this work, we aimed to identify and quantify predictive and causal mechanisms for prostate and SV intrafraction motion, as well as to assess potential reductions in treatment margins given measurements of predictive variables. This study leveraged the intrafractional MR imaging available on MR-Linac to more accurately characterise intrafraction motion. We envisaged that these findings would inform radiotherapy planning across treatment modalities by offering new insights into the physiological and behavioural mechanisms of prostate and SV motion, ultimately supporting safer, more precise, and more individualised care.

2. Materials and methods

2.1. Dataset acquisition

Thirty consenting prostate cancer participants were recruited, with approval from the Townsville Hospital and Health Service Human Research Ethics Committee (HREC/QTHS/71867). Participants were instructed to consume 500 mL of water 20 min before treatment. Rectum size was assessed from the course’s planning computed tomography (CT) and on each daily magnetic resonance imaging (MRI). If excessive, the patients were advised to empty their bowels or were given an enema if needed.

Inclusion criteria included treatment on the Elekta Unity 1.5T MR-Linac (Elekta AB, Stockholm, Sweden), histologically confirmed prostate cancer and aged over 18 years. Each treatment fraction included at least four pelvic MR scans (pre-treatment, planning re-optimisation, pre-beam delivery, and post-beam delivery). A total of 351 T2-weighted MR scans were obtained, using one of the two parameter sets (Supplementary Material Table S1). After excluding 34 outlier scans caused by to pelvic floor muscle tension, mid-fraction bladder emptying, or poor contours (due to ascites in bladder-prostate region), a total of 327 scans remained. Examples of excluded scans are provided in Supplementary Material Figs. S1 and S2. From the remaining dataset, a subset of 58 scans from 12 participants were manually contoured (1x baseline scan for the first fraction of each participant, and randomly selected additional scans from other fractions), by a radiation therapist and reviewed by a radiation oncologist, identifying the prostate, bladder, rectum, and SV. A total of 9 fractions were excluded from the study due to outliers described previously. Results were based on the analysis of 80 fractions and 317 scans.

2.2. Segmentation

An automatic contouring model, nnU-Net, was trained on a subset of manually contoured scans to contour the prostate, bladder, rectum, and SV across the entire dataset [28], using the standard pipeline as per nnU-Net package (version 2.2) [29]. Despite limited training dataset, the model achieved segmentation performance with Dice Similarity Coefficients (DSC) of 0.888 (prostate), 0.817 (SV), 0.833 (rectum), and 0.958 (bladder), comparable to interobserver variability (Supplementary Material Table S2) [28]. For a thorough investigation of the model’s performance with this dataset, we refer the reader to our previous work [28]. Accordingly, all further analyses used automated segmentations, with manual contours reserved for training only. Generated segmentations were manually verified by the primary author to ensure correctness prior to analysis.

2.2.1. Feature extraction

A multidisciplinary team of clinicians consisting of two radiation oncologists, one radiation therapist and one medical physicist, collaborated to conceptualise potential contributing factors influencing prostate and SV intrafraction motion. This led to the development of features outlined in Table 1, which were extracted and assigned from each treatment fraction. Two feature types were defined: “dynamic” (calculated from all session images), and “baseline” (calculated from the pre-treatment scans). While both describe motion, only the baseline features were used for motion prediction.

Table 1.

Extracted Descriptive Features for each treatment session. Features identified with Dyn- are derived from multiple images within the session, and are used to describe motion, whereas Bas- features are derived only from the pre-treatment scan and can be used to predict motion.

ID#	Features	Definition
Dyn-A	Change of bladder volume during fraction	Change in bladder volume from its baseline measurement to each subsequent time point during the treatment fraction, corresponding to when MR images were acquired.
Dyn-B	Change of rectal volume near prostate	Change in rectal cross-sectional area (aligned to the prostate centroid superior-inferior axis) from baseline to each MR imaging time point during the treatment fraction. Area here represents a spatial infinitesimal volume.
Dyn-C	Change of rectal volume near SV	Change in rectal cross-sectional area (aligned to the SV centroid superior-inferior axis) from baseline to each MR imaging time point during the treatment fraction. Area here represents a spatial infinitesimal volume.
Dyn-D	Change of rectal volume near bladder	Change in rectal cross-sectional area (aligned to the bladder centroid superior-inferior axis) from baseline to each MR imaging time point during the treatment fraction. Area here represents a spatial infinitesimal volume.
Bas-A	Baseline Bladder Volume	Volume of the bladder mask in the fraction’s pre-treatment scan.
Bas-B	Rectal volume near SV	Cross-sectional area (to be interpreted as a volume infinitesimal) of an axial slice of the rectum mask, aligned with the superior-interior index of the SV centroid.
Bas-C	Rectal volume near bladder	Cross-sectional area (to be interpreted as a volume infinitesimal) of an axial slice of the rectum mask, aligned with the superior-interior index of the bladder centroid.
Bas-D	Rectal volume near prostate	Cross-sectional area (to be interpreted as a volume infinitesimal) of an axial slice of the rectum mask, aligned with the superior-interior index of the prostate centroid.
Bas-E	Fat Thickness between SV-Bladder	Mean anterior-posterior separation between SV and bladder masks, averaged across axial slices within the SV’s superior-inferior range.
Bas-F	Fat Thickness between SV-Rectum	Mean anterior-posterior separation between SV and rectum masks, averaged across axial slices within the SV’s superior-inferior range.
Bas-G	Fat Thickness between Bladder-Rectum	Mean anterior-posterior separation between bladder and rectum masks, averaged across axial slices within the SV’s superior-inferior range.
Bas-H	Fat Thickness between Prostate-Rectum	Mean anterior-posterior separation between prostate and rectum masks, averaged across axial slices within the prostate’s superior-inferior range.
Bas-I	Hydrogel	Binary labelling on whether hydrogel exist in the MR image.

2.3. Organ intrafraction motion

The Dice Similarity Coefficient (DSC) was utilised as a surrogate measure of intrafractional organ motion. Specifically, it was calculated between the baseline contour and the contours obtained at subsequent timepoints within each treatment fraction. As an overlap metric (0, no overlap, to 1, perfect overlap), it intuitively measured changes due to motion and deformation. In context of radiotherapy, it roughly indicated the proportion of tissues that remain correctly treated under the original plan after motion.

2.4. Direct effect and moderation effect analysis

Univariate linear regression analysis, using ordinary least squares, was performed to assess the direct effect of all features listed in Table 2 on intrafraction organ motion. The statistical significance of each comparison was defined by a p-value of <0.05. Additionally, the variance in motion explained by the feature, via the coefficient of determination (R²).

Table 2.

Statistical analysis on prospective features. * means p-value < 0.05, ** means p-value < 0.005, and *** means p-value < 0.0005. n.s means not significant.

		Prostate Motion (R2)	SV Motion (R2)
ID#	Features
Dyn-A	Change of bladder volume during fraction	***0.11	***0.18
Dyn-B	Change of rectal volume near prostate	n.s	n.s
Dyn-C	Change of rectal volume near SV	n.s	n.s
Dyn-D	Change of rectal volume near bladder	n.s	n.s
Bas-A	Baseline bladder volume	n.s	n.s
Bas-B	Rectal volume near prostate	***0.05	n.s
Bas-C	Rectal volume near SV	n.s	***0.07
Bas-D	Rectal volume near bladder	n.s	**0.04
Bas-E	Fat thickness between prostate-rectum	n.s	n.s
Bas-F	Fat thickness between SV-bladder	*0.01	n.s
Bas-G	Fat thickness between SV-rectum	*** 0.07	**0.04
Bas-H	Fat thickness between bladder-rectum	n.s	*0.01
Bas-I	Hydrogel	n.s	n.s

In addition to quantifying direct effects, we investigated how baseline features influenced the relationship between dynamic features and organ motion. Here, we tested the hypothesis that some patient states, measurable at baseline, may not directly lead to motion but rather may indirectly contribute to motion by moderating the impact of another more directly related variable.

To measure this, we defined the responsiveness of motion to a given feature as the ratio of organ intrafractional motion (as defined by DSC) over each dynamic feature, with the dynamic features themselves being defined as the change from baseline. For each pair of baseline and dynamic features, we identified a binary threshold in the baseline feature to separate the cohort into a responsive and non-responsive set. We then evaluated whether this separation led to an improvement in explained variance in the responsive set. The optimal threshold for this separation was determined by identifying the value that maximised the Cohen’s d between the two distributions of responsiveness. For the responsive group (the group with the higher responsiveness), we determined whether the coefficient of determination (R2) and statistical significance (Bonferroni corrected) between the dynamic feature and organ motion improved compared to the undivided dataset. Baseline features that lead to highly unbalanced groups (less than 10% in one group), to non-significant correlations, or that led to an R² < 10% were excluded.

2.5. Treatment margin

The 95% Hausdorff Distance (HD95) was chosen as a surrogate metric to quantify the margin expansion needed in all directions for uncompromised coverage of the target due to motion. It was calculated as the 95th percentile of the maximum of the minimum distances between 2 boundary contours. In this experiment, HD95 was also used as proxy for estimated motion margin required to be accounted for in the PTV and was calculated for each non-baseline scan.

Similar to the moderation effect analysis, an optimal threshold was determined for each baseline feature to maximise the Cohen’s d between the margin required to capture the motion deformation of prostate or SV in each non-baseline scan (i.e., we separated by each feature into a high margin and low margin group). We used empirical bootstrap estimation to determine whether the 95th percentile margin in each group is statistically significant (p < 0.05).

3. Results

The direct effect of both dynamic and baseline features on SV and prostate intrafraction motion is summarised in Table 2. For dynamic features, bladder filling (Dyn-A) showed a statistically significant influence, explaining up to 11.6% of the prostate and 18.8% of the SV intrafraction motion variability (Supplementary Material Figs. S3 and S4). Rectal filling (Dyn-B, Dyn-C, and Dyn-D) did not show statistical significance. None of the baseline features (i.e., those that could potentially predict motion pre-treatment) exhibited direct linear influence (R² < 0.077). Hydrogel presence (Bas-G) did not have a statistically significant effect on prostate nor SV motion.

As shown in Table 3, among all dynamic features, only organ motion responsivity calculated over the change in bladder filling (Dyn-A) yielded significant moderating effects. Specifically, two baseline features, baseline bladder volume (Bas-A) and fat thickness between prostate-rectum (Bas-E), met the significance criteria. Applying a baseline bladder volume threshold of 190 mL increased the variance by bladder filling from 11.6% to 14.2% in prostate and from 18.8% to 25.8% in SV (Fig. 1). For participants with larger baseline bladder volume, bladder filling tended to expand superiorly and away from the pelvic organs. In contrast, participants with smaller baseline bladder volume exhibited expansion that was confined within the pelvic region. In the 10 fractions with the highest baseline bladder volume, the mean superior bladder expansion was 16.9 mm by the end of fraction. In contrast, in the 10 fractions with the lowest baseline bladder volume, superior expansion by the end of the fraction was just 6.3 mm.

Table 3.

Summary of significant (p < 0.05) and impactful (R² > 0.2) moderator variables and their identified threshold values. The R² columns indicate the proportion of variance in prostate and SV motion explained by a dynamic feature in a direct analysis, after the dataset was partitioned using the identified threshold value. ** means p-value < 0.00625 (after applying the Bonferroni correction).

Prostate motion responsiveness over change in bladder filling (Dyn-A) as mediated by:
Baseline Variable (Moderator)	Baseline Variable Threshold Value (T)	Coefficient of Determination (R2)	Size of Responsive Cohort (n (subset < T))
Baseline Bladder Volume (mL)	189.86	0.142**	146
SV motion responsiveness over change in bladder filling (Dyn-A) as mediated by:
Baseline Variable (Moderator)	Baseline Variable Threshold Value (T)	Coefficient of Determination (R2)	Size of Responsive Cohort (n (subset < T))
Baseline Bladder Volume (mL)	190.39	0.258**	146
Fat Thickness Between Prostate and Rectum (mm)	8.48	0.207**	207

Fig. 1.

Responsiveness of Prostate (Left) and SV (Right) Motion to Change in Bladder Volume (y-axis), against Baseline Bladder Volume (x-axis). The y-axis of this figure represents change of organ motion due to variations in bladder volume during a fraction, while the x-axis denotes the bladder volume at beginning of a fraction. The blue vertical line indicates the identified volume threshold, highlighting that the moderation effect is more pronounced when baseline bladder volume is below this threshold. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Smaller fat thickness between prostate and rectum increased the organ motion responsiveness to bladder filling changes (Fig. 2).

Fig. 2.

Responsiveness of Prostate (Left) and SV (Right) Motion to Change in Bladder Volume (y-axis), against Fat Thickness Between Prostate-Rectum (mm) at Start of Fraction (x-axis). The y-axis of this figure represents change of organ motion due to variations in bladder volume during a fraction, while the x-axis denotes the fat thickness between the prostate and rectum at beginning of a fraction. The blue vertical line indicates the identified threshold for this baseline features. The left panel of this plot suggests a potentially meaningful, but currently inconclusive, moderation effect on prostate motion responsivity. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Among the features examined, only baseline bladder volume (Bas-A) demonstrated a statistically significant relationship with the required margin to account for patient motion (Fig. 3). Bladder volume threshold of 332 mL optimally and significantly separated the required SV margin to 5.8 mm (<332 mL) and 3.5 mm (≥332 mL), compared with 5.7 mm for the whole cohort. The results for the prostate cohort are depicted in the upper panel of Fig. 3: while they did not reach significance, they qualitatively trended towards enabling reduced margins for a pre-treatment bladder volume >300 mL.

Fig. 3.

Prostate & seminal vesicles treatment margin being influenced by baseline bladder volume at start of fraction below and above the determined threshold (vertical dotted line). The horizontal line of this plot denotes the PTV margins required to achieve 95% coverage across all the treatment fractions.

4. Discussion

Using imaging dataset obtained during radiotherapy, this study investigated the predictors and causes of prostate and SV intrafraction motion and explored how measured predictive variables could inform treatment margins. Bladder volume emerged as a key factor, serving both as a cause and a predictor of motion in both organs and influencing the margins required for adequate target coverage.

Additional finding also implied that pre-treatment bladder volume mediates the effect of bladder filling on prostate and SV motion, rather than directly governing the rate of filling itself. When the bladder was inadequately filled at the start of the session, subsequent expansion appears to exert greater mechanical influence on adjacent pelvic organs. Conversely, above the identified threshold, the bladder may already be sufficiently distended, thereby reducing its impact on the surrounding pelvic anatomy. These findings lend support to existing clinical guidelines recommending a fuller bladder to reduce prostate motion, though current evidence remains limited [13], [30], [31], [32]. Importantly, this analysis underscored the potential of a data-driven or personalised approach to motion margin definition, potentially enabling precise treatment margins in patients presenting with an appropriately distended bladder.

Localised rectal filling contributed to both SV and prostate motion. Direct effect analysis indicated that the rectal volume near the prostate (Bas-C) significantly influenced prostate motion, while the rectal volume near to the SV (Bas-D) and near to the bladder (Bas-E) significantly influenced SV motion. Although explanatory power was limited (R² < 0.074), the results suggested that rectal regions adjacent to each organ had the greatest impact on its motion.

This contrasts with prior studies that often treated the rectum as a single structure. Some identified rectal filling as a predictor of prostate motion [9], [33], [34], whereas others, like Nichol et al. [35] and Ogino et al. [12] emphasised rectal gas. Meanwhile, Polat et al. [36] reported a weak correlation (R² = 0.55) between larger prostate motion and rectal volume changes. Oates et al. [11] noted maximum rectal diameter measured on pre-treatment CBCT could predict small prostate intrafraction motion. Frank et al. [21] reported that rectal cross-sectional area correlates with mean A-P interfraction motion of the prostate and SV (R² = 0.303 and 0.3089, respectively).

Separation between prostate and rectum exhibited moderating effect on SV motion. Padhani et al. [37] found that smaller separation between rectum and prostate increased rectal influence on prostate motion. Our analysis of the fat thickness between the prostate and rectum (Bas-E) did not show a direct effect. However, it did exhibit a moderating role in that at determined threshold of 8 mm, it slightly improved the relationship between bladder volume change and SV motion (R2 from 18.8% to 20.7%). This aligns with anatomical expectations, as the rectum’s proximity to both prostate and SV increases the likelihood that rectal deformation and bladder expansion may propagate mechanical forces, influencing organ motion. While the moderating effect was not strong enough to result in an observed reduction in prostate treatment margin, it presents a potentially valuable avenue for future research.

The impact of the presence of a hydrogel spacer on prostate and SV intrafraction motion remained inconclusive due to limited data. The existing literature on the topic is divided, with reports both of negligible [13] and significant, albeit different, effects [23], [24], [25], [37]. Unfortunately, our results do not contribute to the discussion: we found no motion associations with hydrogel presence but conversely were unable to exclude any effect through the TOST analysis, due to only two individuals with hydrogel being recruited.

Baseline bladder volume was a useful predictor of SV treatment margin, consistent with its role in influencing SV motion. Participants with baseline bladder volume below 190 mL experienced increase moderation of SV motion. Similarly, treatment margins increased to 5.8 mm for volumes under 332 mL, compared to 3.5 mm above this threshold. Despite differing thresholds, both analyses reinforced the association between smaller bladder volumes and greater SV motion.

Our SV recommended treatment margins of 5.8 mm for higher-risk and 3.5 mm for lower risk groups are more conservative compared to literature. Dassen et al. [38] recommended a margin of 3.5 mm for adapt-to-shape (ATS) MRgRT workflow. Muinck Keizer et al. [20] reported that an isometric expansion of 5 mm was adequate to cover 99% of the seminal vesicle volume for 95% of the treatment duration, based on coverage probability analysis [39]. Byrne et al. [19] and Sheng et al. [18] similarly supported that 5 mm margins were able to cover 95% of SV intrafraction motion in over 90% of fractions.

Many prior studies utilised the Van Herk’s formula to calculate the theoretical population-based margins [18], [36], [37], [40], [41], incorporating uncertainties such as intrafraction motion, delineation, and registration errors. However, these errors vary by treatment setup which may include imaging modality, real time tracking, hydrogel use, and immobilization techniques [38].

In contrast, our findings offered a complementary, approach to margin calculation. By incorporating predictive baseline features, this methodology allows a personalised and forward-looking strategy for defining margins, thus potentially improving precision and confidence in treatment planning.

Our approach was based on the extraction and analysis of manually extracted features, which were clinically motivated, but also of limited power and variability. Future research could explore data-driven approaches for discovering features directly from the imaging dataset [42]. For example, this could use models such as Variational Autoencoders (VAEs) [43], [44] or Generative Adversarial Networks (GANs) [45] to learn latent representations that capture consistent and informative patterns across a population.

Before adoption of anatomically guided PTV margins, our findings should be validated in a prospective study. For instance, a retrospective treatment planning simulation study could be conducted to evaluate the treatment dose implications with and without the recommended adjusted margins.

A risk in this analysis lied in the imbalance between the number of unique participants (28) and the number of total fractions in this study (80). This raised the possibility that the observed results could be disproportionately influenced by a small subset of participants. However, we believe this is not the case for several reasons.

First, qualitative examination of results did not indicate that any small numbers of participants dominated the trends, as seen in Supplementary Materials Fig. S5, which corresponds to Fig. 1 but represents individual participants by distinct hue and labels.

Second, for the moderation effect analysis assessing how baseline bladder volume influences the relationship between bladder filling and prostate/SV motion, we performed a pseudo-bootstrap analysis, in which each replicate included only a single fraction per participant. Apart from extreme cases where the sample contained exclusively low or high volume bladders, the resulting distributions peaked around 200 mL and values greater than the threshold were associated with lower organ motion responsivity, aligning closely with the reported values.

Lastly, we observed that the inter-patient variance was substantially less than the intra-patient variance for both baseline bladder volume (55 mL vs. 129 mL) and change in bladder filling (11 mL vs. 23 mL). This indicated that the variability in the dataset arises primarily from within participant difference rather than between participants, supporting the reliability of the analysis.

In conclusion, imaging features extracted from MR images obtained during a radiotherapy could serve as predictors and causal factors of organ intrafraction motion. For both prostate and seminal vesicles, our findings indicated that initial bladder volume could be influential in determining the magnitude of intrafraction motion during treatment. These findings may support the prospective determination of treatment margins, enabling more personalised and precise radiotherapy. Lastly, the proposed methodology may motivate further work to derive additional imaging features that better capture and explain the complexity of intrafraction organ motion.

Research data for this article

Due to the sensitive nature of the data collected in this study, participants were assured that their MR imaging data would remain confidential and would not be shared beyond the scope defined by the approved ethics and governance protocols.

Declaration of Generative AI and AI-assisted technologies in the writing process

During the preparation of this work the author(s) used OpenAI’s ChatGPT 4o model to assist with editing and language polishing. All content generated using this tool was subsequently reviewed and revised by the author(s), who take full responsibility for the final content of the publication.

CRediT authorship contribution statement

Febrio Lunardo: Conceptualization, Methodology, Software, Writing – original draft, Formal analysis. Alex Tan: Data curation, Supervision, Writing – review & editing, Resources. Laura Baker: Data curation, Supervision, Writing – review & editing, Resources. John Baines: Data curation, Writing – review & editing, Resources. Timothy Squire: Data curation, Supervision, Writing – review & editing, Resources. Jason A. Dowling: Supervision, Writing – review & editing, Resources. Mostafa Rahimi Azghadi: Supervision, Writing – review & editing, Resources. Ashley G. Gillman: Conceptualization, Methodology, Software, Supervision, Writing – review & editing, Resources.

Funding

Febrio Lunardo is supported by the Research Training Program (RTP) provided by the Department of Education of the Australian Federal Government and a Commonwealth Scientific and Industrial Research Organisation (CSIRO) Top-up Scholarship.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: The authors have no relevant financial or non-financial interests to disclose.

Appendix A. Supplementary data

The following are the Supplementary data to this article: