Uncertainty estimation using a 3D probabilistic U-Net for segmentation with small radiotherapy clinical trial datasets.

BACKGROUND AND OBJECTIVES: Bio-medical image segmentation models typically attempt to predict one segmentation that resembles a ground-truth structure as closely as possible. However, as medical images are not perfect representations of anatomy, obtaining this ground truth is not possible. A surrogate commonly used is to have multiple expert observers define the same structure for a dataset. When multiple observers define the same structure on the same image there can be significant differences depending on the structure, image quality/modality and the region being defined. It is often desirable to estimate this type of aleatoric uncertainty in a segmentation model to help understand the region in which the true structure is likely to be positioned. Furthermore, obtaining these datasets is resource intensive so training such models using limited data may be required. With a small dataset size, differing patient anatomy is likely not well represented causing epistemic uncertainty which should also be estimated so it can be determined for which cases the model is effective or not. METHODS: We use a 3D probabilistic U-Net to train a model from which several segmentations can be sampled to estimate the range of uncertainty seen between multiple observers. To ensure that regions where observers disagree most are emphasised in model training, we expand the Generalised Evidence Lower Bound (ELBO) with a Constrained Optimisation (GECO) loss function with an additional contour loss term to give attention to this region. Ensemble and Monte-Carlo dropout (MCDO) uncertainty quantification methods are used during inference to estimate model confidence on an unseen case. We apply our methodology to two radiotherapy clinical trial datasets, a gastric cancer trial (TOPGEAR, TROG 08.08) and a post-prostatectomy prostate cancer trial (RAVES, TROG 08.03). Each dataset contains only 10 cases each for model development to segment the clinical target volume (CTV) which was defined by multiple observers on each case. An additional 50 cases are available as a hold-out dataset for each trial which had only one observer define the CTV structure on each case. Up to 50 samples were generated using the probabilistic model for each case in the hold-out dataset. To assess performance, each manually defined structure was matched to the closest matching sampled segmentation based on commonly used metrics. RESULTS: The TOPGEAR CTV model achieved a Dice Similarity Coefficient (DSC) and Surface DSC (sDSC) of 0.7 and 0.43 respectively with the RAVES model achieving 0.75 and 0.71 respectively. Segmentation quality across cases in the hold-out datasets was variable however both the ensemble and MCDO uncertainty estimation approaches were able to accurately estimate model confidence with a p-value < 0.001 for both TOPGEAR and RAVES when comparing the DSC using the Pearson correlation coefficient. CONCLUSIONS: We demonstrated that training auto-segmentation models which can estimate aleatoric and epistemic uncertainty using limited datasets is possible. Having the model estimate prediction confidence is important to understand for which unseen cases a model is likely to be useful.

Evaluating the relationship between contouring variability and modelled treatment outcome for prostate bed radiotherapy.

Objectives.Contouring similarity metrics are often used in studies of inter-observer variation and automatic segmentation but do not provide an assessment of clinical impact. This study focused on post-prostatectomy radiotherapy and aimed to (1) identify if there is a relationship between variations in commonly used contouring similarity metrics and resulting dosimetry and (2) identify the variation in clinical target volume (CTV) contouring that significantly impacts dosimetry.Approach.The study retrospectively analysed CT scans of 10 patients from the TROG 08.03 RAVES trial. The CTV, rectum, and bladder were contoured independently by three experienced observers. Using these contours reference simultaneous truth and performance level estimation (STAPLE) volumes were established. Additional CTVs were generated using an atlas algorithm based on a single benchmark case with 42 manual contours. Volumetric-modulated arc therapy (VMAT) treatment plans were generated for the observer, atlas, and reference volumes. The dosimetry was evaluated using radiobiological metrics. Correlations between contouring similarity and dosimetry metrics were calculated using Spearman coefficient (Γ). To access impact of variations in planning target volume (PTV) margin, the STAPLE PTV was uniformly contracted and expanded, with plans created for each PTV volume. STAPLE dose-volume histograms (DVHs) were exported for plans generated based on the contracted/expanded volumes, and dose-volume metrics assessed.Mainresults. The study found no strong correlations between the considered similarity metrics and modelled outcomes. Moderate correlations (0.5 <Γ< 0.7) were observed for Dice similarity coefficient, Jaccard, and mean distance to agreement metrics and rectum toxicities. The observations of this study indicate a tendency for variations in CTV contraction/expansion below 5 mm to result in minor dosimetric impacts.Significance. Contouring similarity metrics must be used with caution when interpreting them as indicators of treatment plan variation. For post-prostatectomy VMAT patients, this work showed variations in contours with an expansion/contraction of less than 5 mm did not lead to notable dosimetric differences, this should be explored in a larger dataset to assess generalisability.

Assessment of intrafraction motion and its dosimetric impact on prostate radiotherapy using an in-house developed position monitoring system.

Purpose: To implement an in-house developed position monitoring software, SeedTracker, for conventional fractionation prostate radiotherapy, and study the effect on dosimetric impact and intrafraction motion. Methods: Thirty definitive prostate radiotherapy patients with implanted fiducial markers were included in the study. All patients were treated with VMAT technique and plans were generated using the Pinnacle planning system using the 6MV beam model for Elekta linear accelerator. The target dose of 60 Gy in 20 fractions was prescribed for 29 of 30 patients, and one patient was treated with the target dose of 78 Gy in 39 fractions. The SeedTracker position monitoring system, which uses the x-ray images acquired during treatment delivery in the Elekta linear accelerator and associated XVI system, was used for online prostate position monitoring. The position tolerance for online verification was progressively reduced from 5 mm, 4 mm, and to 3 mm in 10 patient cohorts to effectively manage the treatment interruptions resulting from intrafraction motion in routine clinical practice. The delivered dose to target volumes and organs at risk in each of the treatment fractions was assessed by incorporating the observed target positions into the original treatment plan. Results: In 27 of 30 patients, at least one gating event was observed, with a total of 177 occurrences of position deviation detected in 146 of 619 treatment fractions. In 5 mm, 4 mm, and 3 mm position tolerance cohorts, the position deviations were observed in 13%, 24%, and 33% of treatment fractions, respectively. Overall, the mean (range) deviation of -0.4 (-7.2 to 5.3) mm, -0.9 (-6.1 to 15.6) mm, and -1.7 (-7.0 to 6.1) mm was observed in Left-Right, Anterior-Posterior, and Superior-Inferior directions, respectively. The prostate CTV D99 would have been reduced by a maximum value of 1.3 Gy compared to the planned dose if position deviations were uncorrected, but with corrections, it was 0.3 Gy. Similarly, PTV D98 would have been reduced by a maximum value of 7.6 Gy uncorrected, with this difference reduced to 2.2 Gy with correction. The V60 to the rectum increased by a maximum of 1.0% uncorrected, which was reduced to 0.5%. Conclusion: Online target position monitoring for conventional fractionation prostate radiotherapy was successfully implemented on a standard Linear accelerator using an in-house developed position monitoring software, with an improvement in resultant dose to prostate target volume.

Clinical target volume delineation quality assurance for MRI-guided prostate radiotherapy using deep learning with uncertainty estimation.

BACKGROUND AND PURPOSE: Previous studies on automatic delineation quality assurance (QA) have mostly focused on CT-based planning. As MRI-guided radiotherapy is increasingly utilized in prostate cancer treatment, there is a need for more research on MRI-specific automatic QA. This work proposes a clinical target volume (CTV) delineation QA framework based on deep learning (DL) for MRI-guided prostate radiotherapy. MATERIALS AND METHODS: The proposed workflow utilized a 3D dropblock ResUnet++ (DB-ResUnet++) to generate multiple segmentation predictions via Monte Carlo dropout which were used to compute an average delineation and area of uncertainty. A logistic regression (LR) classifier was employed to classify the manual delineation as pass or discrepancy based on the spatial association between the manual delineation and the network's outputs. This approach was evaluated on a multicentre MRI-only prostate radiotherapy dataset and compared with our previously published QA framework based on AN-AG Unet. RESULTS: The proposed framework achieved an area under the receiver operating curve (AUROC) of 0.92, a true positive rate (TPR) of 0.92 and a false positive rate of 0.09 with an average processing time per delineation of 1.3 min. Compared with our previous work using AN-AG Unet, this method generated fewer false positive detections at the same TPR with a much faster processing speed. CONCLUSION: To the best of our knowledge, this is the first study to propose an automatic delineation QA tool using DL with uncertainty estimation for MRI-guided prostate radiotherapy, which can potentially be used for reviewing prostate CTV delineation in multicentre clinical trials.

Changes in serial multiparametric MRI and FDG-PET/CT functional imaging during radiation therapy can predict treatment response in patients with head and neck cancer.

OBJECTIVES: To test if tumour changes measured using combination of diffusion-weighted imaging (DWI) MRI and FDG-PET/CT performed serially during radiotherapy (RT) in mucosal head and neck carcinoma can predict treatment response. METHODS: Fifty-five patients from two prospective imaging biomarker studies were analysed. FDG-PET/CT was performed at baseline, during RT (week 3), and post RT (3 months). DWI was performed at baseline, during RT (weeks 2, 3, 5, 6), and post RT (1 and 3 months). The ADCmean from DWI and FDG-PET parameters SUVmax, SUVmean, metabolic tumour volume (MTV), and total lesion glycolysis (TLG) were measured. Absolute and relative change (%∆) in DWI and PET parameters were correlated to 1-year local recurrence. Patients were categorised into favourable, mixed, and unfavourable imaging response using optimal cut-off (OC) values of DWI and FDG-PET parameters and correlated to local control. RESULTS: The 1-year local, regional, and distant recurrence rates were 18.2% (10/55), 7.3% (4/55), and 12.7% (7/55), respectively. ∆Week 3 ADCmean (AUC 0.825, p = 0.003; OC ∆ > 24.4%) and ∆MTV (AUC 0.833, p = 0.001; OC ∆ > 50.4%) were the best predictors of local recurrence. Week 3 was the optimal time point for assessing DWI imaging response. Using a combination of ∆ADCmean and ∆MTV improved the strength of correlation to local recurrence (p ≤ 0.001). In patients who underwent both week 3 MRI and FDG-PET/CT, significant differences in local recurrence rates were seen between patients with favourable (0%), mixed (17%), and unfavourable (78%) combined imaging response. CONCLUSIONS: Changes in mid-treatment DWI and FDG-PET/CT imaging can predict treatment response and could be utilised in the design of future adaptive clinical trials. CLINICAL RELEVANCE STATEMENT: Our study shows the complementary information provided by two functional imaging modalities for mid-treatment response prediction in patients with head and neck cancer. KEY POINTS: •FDG-PET/CT and DWI MRI changes in tumour during radiotherapy in head and neck cancer can predict treatment response. •Combination of FDG-PET/CT and DWI parameters improved correlation to clinical outcome. •Week 3 was the optimal time point for DWI MRI imaging response assessment.

Mid-treatment 18F-FDG PET imaging changes in parotid gland correlates to radiation-induced xerostomia.

BACKGROUND: The aim of this study was to measure functional changes in parotid glands using mid-treatment FDG-PET/CT and correlate early imaging changes to subsequent xerostomia in mucosal head and neck squamous cell carcinoma patients undergoing radiotherapy. MATERIALS AND METHODS: 56 patients from two prospective imaging biomarker studies underwent FDG-PET/CT at baseline and during radiotherapy (week 3). Both parotid glands were volumetrically delineated at each time point. PET parameter SUVmedian were calculated for ipsilateral and contralateral parotid glands. Absolute and relative change (Δ) in SUVmedian were correlated to moderate-severe xerostomia (CTCAE grade ≥ 2) at 6 months. Four predictive models were subsequently created using multivariate logistic regression using clinical and radiotherapy planning parameters. Model performance was calculated using ROC analysis and compared using Akaike information criterion (AIC) RESULTS: 29 patients (51.8%) developed grade ≥ 2 xerostomia. Compared to baseline, there was an increase in SUVmedian at week 3 in ipsilateral (8.4%) and contralateral (5.5%) parotid glands. Increase in ipsilateral parotid Δ SUVmedian (p = 0.04) and contralateral mean parotid dose (p = 0.04) were correlated to xerostomia. The reference 'clinical' model correlated to xerostomia (AUC 0.667, AIC 70.9). Addition of ipsilateral parotid Δ SUVmedian to the clinical model resulted in the highest correlation to xerostomia (AUC 0.777, AIC 65.4). CONCLUSION: Our study shows functional changes occurring in the parotid gland early during radiotherapy. We demonstrate that integration of baseline and mid-treatment FDG-PET/CT changes in the parotid gland with clinical factors has the potential to improve xerostomia risk prediction which could be utilised for personalised head and neck radiotherapy.

Impact of tumour region of interest delineation method for mid-treatment FDG-PET response prediction in head and neck squamous cell carcinoma undergoing radiotherapy.

Background: The aim of this study was to evaluate the impact of tumour region of interest (ROI) delineation method on mid-treatment 18F-fluoro-2-deoxy-D-glucose positron emission tomography (FDG-PET) response prediction in mucosal head and neck squamous cell carcinoma during radiotherapy. Methods: A total of 52 patients undergoing definitive radiotherapy with or without systemic therapy from two prospective imaging biomarker studies were analysed. FDG-PET was performed at baseline and during radiotherapy (week 3). Primary tumour was delineated using a fixed SUV 2.5 threshold (MTV2.5), relative threshold (MTV40%) and a gradient based segmentation method (PET Edge). PET parameters SUVmax, SUVmean, metabolic tumour volume (MTV) and total lesion glycolysis (TLG) were calculated using different ROI methods. Absolute and relative change (∆) in PET parameters were correlated to 2-year locoregional recurrence. Strength of correlation was tested using receiver operator characteristic analysis using area under the curve (AUC). Response was categorized using optimal cut-off (OC) values. Correlation and agreement between different ROI methods was determined using Bland-Altman analysis. Results: A significant difference in SUVmean, MTV and TLG values were noted between ROI delineation methods. When measuring relative change at week 3, a greater agreement was seen between PET Edge and MTV2.5 methods with average difference in ∆SUVmax, ∆SUVmean, ∆MTV and ∆TLG of 0.0%, 3.6%, 10.3% and 13.6% respectively. A total of 12 patients (22.2%) experienced locoregional recurrence. ∆MTV using PET Edge was the best predictor of locoregional recurrence (AUC =0.761, 95% CI: 0.573-0.948, P=0.001; OC ∆>50%). The corresponding 2-year locoregional recurrence rate was 7% vs. 35%, P=0.001. Conclusions: Our findings suggest that it is preferable to use gradient based method to assess volumetric tumour response during radiotherapy and offers advantage in predicting treatment outcomes compared with threshold-based methods. This finding requires further validation and can assist in future response-adaptive clinical trials.

Realistic CT data augmentation for accurate deep-learning based segmentation of head and neck tumors in kV images acquired during radiation therapy.

BACKGROUND: Using radiation therapy (RT) to treat head and neck (H&N) cancers requires precise targeting of the tumor to avoid damaging the surrounding healthy organs. Immobilisation masks and planning target volume margins are used to attempt to mitigate patient motion during treatment, however patient motion can still occur. Patient motion during RT can lead to decreased treatment effectiveness and a higher chance of treatment related side effects. Tracking tumor motion would enable motion compensation during RT, leading to more accurate dose delivery. PURPOSE: The purpose of this paper is to develop a method to detect and segment the tumor in kV images acquired during RT. Unlike previous tumor segmentation methods for kV images, in this paper, a process for generating realistic and synthetic CT deformations was developed to augment the training data and make the segmentation method robust to patient motion. Detecting the tumor in 2D kV images is a necessary step toward 3D tracking of the tumor position during treatment. METHOD: In this paper, a conditional generative adversarial network (cGAN) is presented that can detect and segment the gross tumor volume (GTV) in kV images acquired during H&N RT. Retrospective data from 15 H&N cancer patients obtained from the Cancer Imaging Archive were used to train and test patient-specific cGANs. The training data consisted of digitally reconstructed radiographs (DRRs) generated from each patient's planning CT and contoured GTV. Training data was augmented by using synthetically deformed CTs to generate additional DRRs (in total 39 600 DRRs per patient or 25 200 DRRs for nasopharyngeal patients) containing realistic patient motion. The method for deforming the CTs was a novel deformation method based on simulating head rotation and internal tumor motion. The testing dataset consisted of 1080 DRRs for each patient, obtained by deforming the planning CT and GTV at different magnitudes to the training data. The accuracy of the generated segmentations was evaluated by measuring the segmentation centroid error, Dice similarity coefficient (DSC) and mean surface distance (MSD). This paper evaluated the hypothesis that when patient motion occurs, using a cGAN to segment the GTV would create a more accurate segmentation than no-tracking segmentations from the original contoured GTV, the current standard-of-care. This hypothesis was tested using the 1-tailed Mann-Whitney U-test. RESULTS: The magnitude of our cGAN segmentation centroid error was (mean ± standard deviation) 1.1 ± 0.8 mm and the DSC and MSD values were 0.90 ± 0.03 and 1.6 ± 0.5 mm, respectively. Our cGAN segmentation method reduced the segmentation centroid error (p < 0.001), and MSD (p = 0.031) when compared to the no-tracking segmentation, but did not significantly increase the DSC (p = 0.294). CONCLUSIONS: The accuracy of our cGAN segmentation method demonstrates the feasibility of this method for H&N cancer patients during RT. Accurate tumor segmentation of H&N tumors would allow for intrafraction monitoring methods to compensate for tumor motion during treatment, ensuring more accurate dose delivery and enabling better H&N cancer patient outcomes.

Open-source, fully-automated hybrid cardiac substructure segmentation: development and optimisation.

Radiotherapy for thoracic and breast tumours is associated with a range of cardiotoxicities. Emerging evidence suggests cardiac substructure doses may be more predictive of specific outcomes, however, quantitative data necessary to develop clinical planning constraints is lacking. Retrospective analysis of patient data is required, which relies on accurate segmentation of cardiac substructures. In this study, a novel model was designed to deliver reliable, accurate, and anatomically consistent segmentation of 18 cardiac substructures on computed tomography (CT) scans. Thirty manually contoured CT scans were included. The proposed multi-stage method leverages deep learning (DL), multi-atlas mapping, and geometric modelling to automatically segment the whole heart, cardiac chambers, great vessels, heart valves, coronary arteries, and conduction nodes. Segmentation performance was evaluated using the Dice similarity coefficient (DSC), mean distance to agreement (MDA), Hausdorff distance (HD), and volume ratio. Performance was reliable, with no errors observed and acceptable variation in accuracy between cases, including in challenging cases with imaging artefacts and atypical patient anatomy. The median DSC range was 0.81-0.93 for whole heart and cardiac chambers, 0.43-0.76 for great vessels and conduction nodes, and 0.22-0.53 for heart valves. For all structures the median MDA was below 6 mm, median HD ranged 7.7-19.7 mm, and median volume ratio was close to one (0.95-1.49) for all structures except the left main coronary artery (2.07). The fully automatic algorithm takes between 9 and 23 min per case. The proposed fully-automatic method accurately delineates cardiac substructures on radiotherapy planning CT scans. Robust and anatomically consistent segmentations, particularly for smaller structures, represents a major advantage of the proposed segmentation approach. The open-source software will facilitate more precise evaluation of cardiac doses and risks from available clinical datasets.

Standardising Breast Radiotherapy Structure Naming Conventions: A Machine Learning Approach.

In progressing the use of big data in health systems, standardised nomenclature is required to enable data pooling and analyses. In many radiotherapy planning systems and their data archives, target volumes (TV) and organ-at-risk (OAR) structure nomenclature has not been standardised. Machine learning (ML) has been utilised to standardise volumes nomenclature in retrospective datasets. However, only subsets of the structures have been targeted. Within this paper, we proposed a new approach for standardising all the structures nomenclature by using multi-modal artificial neural networks. A cohort consisting of 1613 breast cancer patients treated with radiotherapy was identified from Liverpool & Macarthur Cancer Therapy Centres, NSW, Australia. Four types of volume characteristics were generated to represent each target and OAR volume: textual features, geometric features, dosimetry features, and imaging data. Five datasets were created from the original cohort, the first four represented different subsets of volumes and the last one represented the whole list of volumes. For each dataset, 15 sets of combinations of features were generated to investigate the effect of using different characteristics on the standardisation performance. The best model reported 99.416% classification accuracy over the hold-out sample when used to standardise all the nomenclatures in a breast cancer radiotherapy plan into 21 classes. Our results showed that ML based automation methods can be used for standardising naming conventions in a radiotherapy plan taking into consideration the inclusion of multiple modalities to better represent each volume.

Repeatability and reproducibility of magnetic resonance imaging-based radiomic features in rectal cancer.

Purpose: Radiomics of magnetic resonance images (MRIs) in rectal cancer can non-invasively characterize tumor heterogeneity with potential to discover new imaging biomarkers. However, for radiomics to be reliable, the imaging features measured must be stable and reproducible. The aim of this study is to quantify the repeatability and reproducibility of MRI-based radiomic features in rectal cancer. Approach: An MRI radiomics phantom was used to measure the longitudinal repeatability of radiomic features and the impact of post-processing changes related to image resolution and noise. Repeatability measurements in rectal cancers were also quantified in a cohort of 10 patients with test-retest imaging among two observers. Results: We found that many radiomic features, particularly from texture classes, were highly sensitive to changes in image resolution and noise. About 49% of features had coefficient of variations ≤ 10 % in longitudinal phantom measurements. About 75% of radiomic features in in vivo test-retest measurements had an intraclass correlation coefficient of ≥ 0.8 . We saw excellent interobserver agreement with mean Dice similarity coefficient of 0.95 ± 0.04 for test and retest scans. Conclusions: The results of this study show that even when using a consistent imaging protocol many radiomic features were unstable. Therefore, caution must be taken when selecting features for potential imaging biomarkers.

Development of a vendor neutral MRI distortion quality assurance workflow.

With the utilization of magnetic resonance (MR) imaging in radiotherapy increasing, routine quality assurance (QA) of these systems is necessary. The assessment of geometric distortion in images used for radiotherapy treatment planning needs to be quantified and monitored over time. This work presents an adaptable methodology for performing routine QA for systematic MRI geometric distortion. A software tool and compatible protocol (designed to work with any CT and MR compatible phantom on any scanner) were developed to quantify geometric distortion via deformable image registration. The MR image is deformed to the CT, generating a deformation field, which is sampled, quantifying geometric distortion as a function of distance from scanner isocenter. Configurability of the QA tool was tested, and results compared to those provided from commercial solutions. Registration accuracy was investigated by repeating the deformable registration step on the initial deformed MR image to define regions with residual distortions. The geometric distortion of four clinical systems was quantified using the customisable QA method presented. Maximum measured distortions varied from 2.2 to 19.4 mm (image parameter and sampling volume dependent). The workflow was successfully customized for different phantom configurations and volunteer imaging studies. Comparison to a vendor supplied solution showed good agreement in regions where the two procedures were sampling the same imaging volume. On a large field of view phantom across various scanners, the QA tool accurately quantified geometric distortions within 17-22 cm from scanner isocenter. Beyond these regions, the geometric integrity of images in clinical applications should be considered with a higher degree of uncertainty due to increased gradient nonlinearity and B0 inhomogeneity. This tool has been successfully integrated into routine QA of the MRI scanner utilized for radiotherapy within our department. It enables any low susceptibility MR-CT compatible phantom to quantify the geometric distortion on any MRI scanner with a configurable, user friendly interface for ease of use and consistency in data collection and analysis.

Optimal and actual rates of Stereotactic Ablative Body Radiotherapy (SABR) utilisation for primary lung cancer in Australia.

Background and purpose: Radiotherapy utilisation rates considerably vary across different countries and service providers, highlighting the need to establish reliable benchmarks against which utilisation rates can be assessed. Here, optimal utilisation rates of Stereotactic Ablative Body Radiotherapy (SABR) for lung cancer are estimated and compared against actual utilisation rates to identify potential shortfalls in service provision. Materials and Methods: An evidence-based optimal utilisation model was constructed after reviewing practice guidelines and identifying indications for lung SABR based on the best available evidence. The proportions of patients likely to develop each indication were obtained, whenever possible, from Australian population-based studies. Sensitivity analysis was performed to account for variations in epidemiological data. Practice pattern studies were reviewed to obtain actual utilisation rates. Results: A total of 6% of all lung cancer patients were estimated to optimally require SABR at least once during the course of their illness (95% CI: 4-6%). Optimal utilisation rates were estimated to be 32% for stage I and 10% for stage II NSCLC. Actual utilisation rates for stage I NSCLC varied between 6 and 20%. For patients with inoperable stage I, 27-74% received SABR compared to the estimated optimal rate of 82%. Conclusion: The estimated optimal SABR utilisation rates for lung cancer can serve as useful benchmarks to highlight gaps in service delivery and help plan for more adequate and efficient provision of care. The model can be easily modified to determine optimal utilisation rates in other populations or updated to reflect any changes in practice guidelines or epidemiological data.

Automated post-operative brain tumour segmentation: A deep learning model based on transfer learning from pre-operative images.

Automated brain tumour segmentation from post-operative images is a clinically relevant yet challenging problem. In this study, an automated method for segmenting brain tumour into its subregions has been developed. The dataset consists of multimodal post-operative brain scans (T1 MRI, post-Gadolinium T1 MRI, and T2-FLAIR images) of 15 patients who were treated with post-operative radiation therapy, along with manual annotations of their tumour subregions. A 3D densely-connected U-net was developed for segmentation of brain tumour regions and extensive experiments were conducted to enhance model accuracy. A model was initially developed using the publicly available BraTS dataset consisting of pre-operative brain scans. This model achieved Dice Scores of 0.90, 0.83 and 0.78 for predicting whole tumour, tumour core, and enhancing tumour subregions when tested on BraTS20 blind validation dataset. The acquired knowledge from BraTS was then transferred to the local dataset. For augmentation purpose, the local dataset was registered to a dataset of MRI brain scans of healthy subjects. To improve the robustness of the model and enhance its accuracy, ensemble learning was used to combine the outputs of all the trained models. Even though the size of the dataset is very small, the final model can segment brain tumours with a high Dice Score of 0.83, 0.77 and 0.60 for whole tumour, tumour core and enhancing core respectively.

Variability of gross tumour volume delineation: MRI and CT based tumour and lymph node delineation for lung radiotherapy.

PURPOSE: To compare gross tumour volume (GTV) delineation of lung cancer on magnetic resonance imaging (MRI) and positron emission tomography (PET) versus computed tomography (CT) and PET. METHODS: Three experienced thoracic radiation oncologists delineated GTVs on twenty-six patients with lung cancer, based on CT registered to PET, T2-weighted MRI registered to PET and T1-weighted MRI registered with PET. All observers underwent education on reviewing T1 and T2 images along with guidance on window and level setup. Interobserver and intermodality variation was performed based on dice similarity coefficient (DSC), Hausdorff distance (HD), and average Hausdorff distance (AvgHD) metrics. To compute interobserver variability (IOV) a simultaneous truth and performance level estimation (STAPLE) volume for each image modality was used as reference volume. For intermodality analysis, each observers CT based primary and nodal GTV was used as reference volume. RESULTS: A mean DSC of 0.9 across all observers for primary GTV (GTVp) and a DSC of >0.7 for nodal GTV (GTVn) was demonstrated for IOV. Mean T2 and T1 GTVp and GTVn were smaller than CT GTVp and GTVn but the difference in volume between modalities was not statistically significant. Significant difference (p < 0.01) for GTVp and GTVn was found between T2 and T1 GTVp and GTVn compared to CT GTVp and GTVn based on DSC metrics. Large variation in volume similarity was noted based on HD of up-to 5.4 cm for observer volumes compared to STAPLE volume. CONCLUSION: Interobserver variability in GTV delineation was similar for MRI and PET versus CT and PET. The significant difference between MRI compared to CT delineated volumes needs to be further explored.

Automatic radiotherapy delineation quality assurance on prostate MRI with deep learning in a multicentre clinical trial.

Volume delineation quality assurance (QA) is particularly important in clinical trial settings where consistent protocol implementation is required, as outcomes will affect future as well current patients. Currently, where feasible, this is conducted manually, which is time consuming and resource intensive. Although previous studies mostly focused on automating delineation QA on CT, magnetic resonance imaging (MRI) is being increasingly used in radiotherapy treatment. In this work, we propose to perform automatic delineation QA on prostate MRI for both the clinical target volume (CTV) and organs-at-risk (OARs) by using delineations generated by 3D Unet variants as benchmarks for QA. These networks were trained on a small gold standard atlas set and applied on a multicentre radiotherapy clinical trial dataset to generate benchmark delineations. Then, a QA stage was designed to recommend 'pass', 'minor correction' and 'major correction' for each manual delineation in the trial set by thresholding its Dice similarity coefficient to the network generated delineation. Among all 3D Unet variants explored, the Unet with anatomical gates in an AtlasNet architecture performed the best in delineation QA, achieving an area under the receiver operating characteristics curve of 0.97, 0.92, 0.89 and 0.97 for identifying unacceptable (major correction) delineations with a sensitivity of 0.93, 0.73, 0.74 and 0.90 at a specificity of 0.93, 0.86, 0.86 and 0.95 for bladder, prostate CTV, rectum and gel spacer respectively. To the best of our knowledge, this is the first study to propose automated delineation QA for a multicentre radiotherapy clinical trial with treatment planning MRI. The methods proposed in this work can potentially improve the accuracy and consistency of CTV and OAR delineation in radiotherapy treatment planning.

A review of medical image data augmentation techniques for deep learning applications.

Research in artificial intelligence for radiology and radiotherapy has recently become increasingly reliant on the use of deep learning-based algorithms. While the performance of the models which these algorithms produce can significantly outperform more traditional machine learning methods, they do rely on larger datasets being available for training. To address this issue, data augmentation has become a popular method for increasing the size of a training dataset, particularly in fields where large datasets aren't typically available, which is often the case when working with medical images. Data augmentation aims to generate additional data which is used to train the model and has been shown to improve performance when validated on a separate unseen dataset. This approach has become commonplace so to help understand the types of data augmentation techniques used in state-of-the-art deep learning models, we conducted a systematic review of the literature where data augmentation was utilised on medical images (limited to CT and MRI) to train a deep learning model. Articles were categorised into basic, deformable, deep learning or other data augmentation techniques. As artificial intelligence models trained using augmented data make their way into the clinic, this review aims to give an insight to these techniques and confidence in the validity of the models produced.

Assessing tumor centrality in lung stereotactic ablative body radiotherapy (SABR): the effects of variations in bronchial tree delineation and potential for automated methods.

Accurate delineation of the proximal bronchial tree (PBT) is crucial for appropriate assessment of lung tumor centrality and choice of Stereotactic Ablative Body Radiotherapy (SABR) dose prescription. Here, we investigate variabilities in manual PBT delineation and their potential to influence assessing lesion centrality. A fully automatic, intensity-based tool for PBT contouring and measuring distance to the target is also described. This retrospective analysis included a total of 61 patients treated with lung SABR. A subset of 41 patients was used as a training dataset, containing clinical PBT contour and additional subsequently generated manual contours. The tool was optimized and compared against manual contours in terms of volume, distance to the target and various overlap/similarity metrics. The remaining 20 patients were used as a validation dataset to investigate the dosimetric effects of variations between manual and automatic PBT contours. Considerable interobserver variability was observed, particularly in identifying the superior and inferior borders of the PBT. Automatic PBT contours were comparable to manual contours with average Dice of 0.63 to 0.79 and mean distance to agreement of 1.78 to 3.34 mm. No significant differences in dosimetric parameters were found between automatically and manually generated contours. A moderate negative correlation was found between PBT maximum dose and distance to the lesion (p < 0.05). Variability in manual PBT delineation may result in inconsistent assessment of tumor centrality. Automatic contouring can help standardize clinical practice, support investigations into the link between SABR outcomes and lesion proximity to central airways and the development of predictive toxicity models that incorporate precise measurements of tumor location in relation to high-risk organs.

Data-Driven Modeling of Intracellular Auxin Fluxes Indicates a Dominant Role of the ER in Controlling Nuclear Auxin Uptake.

In plants, the phytohormone auxin acts as a master regulator of developmental processes and environmental responses. The best characterized process in the auxin regulatory network occurs at the subcellular scale, wherein auxin mediates signal transduction into transcriptional programs by triggering the degradation of Aux/IAA transcriptional repressor proteins in the nucleus. However, whether and how auxin movement between the nucleus and the surrounding compartments is regulated remain elusive. Using a fluorescent auxin analog, we show that its diffusion into the nucleus is restricted. By combining mathematical modeling with time course assays on auxin-mediated nuclear signaling and quantitative phenotyping in single plant cell systems, we show that ER-to-nucleus auxin flux represents a major subcellular pathway to directly control nuclear auxin levels. Our findings propose that the homeostatically regulated auxin pool in the ER and ER-to-nucleus auxin fluxes underpin auxin-mediated downstream responses in plant cells.