Young Investigators Symposium - Lombardy/Umbria
le 6 juin 2024 from 13h00 CDT to 14h30 CDT
Scientific Session – Young Investigators Symposium
Thursday, June 6, 2024, 1:00-14:30
Scientific Session YIS – Presentation 1
Incorporation of gold nanoparticles as radiosensitizing agents into brachytherapy and conventional radiotherapy
Daniel Cecchi, Devika Chithrani, Mehran Goharian, Wayne Beckham
University of Victoria
Purpose: To determine the applicability of gold nanoparticles (GNPs) as radiosensitizers for high-dose-rate (HDR) brachytherapy compared to conventional radiotherapy.
Methods: A novel phantom was constructed from blocks of Solid Water, enabling cell irradiations from an Ir192 HDR source and 6MV linear accelerator (LINAC). HeLa and PC3 cell lines were used in all experiments. GNPs of concentration 243.9μg/mL were functionalized with polyethylene glycol and peptide-containing integrin binding domain (RGD) and incubated into the cells (10mg/mL) 24 hours before irradiation to a therapeutically relevant dose of 2Gy. The therapeutic outcome was quantified via DNA double-strand break (DSB) and proliferation assays.
Results: EBT3 film dosimetry confirmed a uniform dose delivery to within ±7% and ±3% of the 2Gy prescription for HDR and LINAC photons, respectively. From Ir192 irradiations relative to control for HeLa and PC3 cells, GNPs increased DSB foci by 92% (p=0.002) and 17% (p=0.25) and reduced proliferation rates two-three days post-irradiation by 15% (p=0.002) and 43% (p<0.0001) for respectively. Compared to MV irradiations for HeLa and PC3 cells, Ir192 induced 59% (p=0.003) and 62% (p=0.004) more DSB foci per nucleus, respectively. 2-3 days post-irradiation, proliferation rates from Ir192 irradiations were reduced for GNP-incubated HeLa cells by 24% (p<0.01) and GNP-incubated PC3 cells by 11% (p=0.002) relative to MV irradiations.
Conclusion: We successfully developed a novel solid water phantom for irradiation from an HDR Ir192 source and a LINAC beam. Compared to MV photons, using GNPs with brachytherapy is promising, given their greater photoelectric cross-section at lower energies, and elicits further investigation.
Scientific Session YIS – Presentation 2
Towards Consistent All-Organ Segmentation in MRI-based Cervical Brachytherapy: Deep Learning-based Clinician-Ready Autocontours with Large Multi-Applicator Dataset.
Patricia Oliva, Shrimanti, Ghos, Fleur Huang, Kumaradevan Punithakumar, Ericka Wiebe, Julie Cuartero, Pierre Boulanger, Jihyun Yun, Geetha Menon
University of Alberta
Purpose: To develop a deep learning (DL)-driven, clinical translation-ready model for simultaneous autosegmentation of the four core organs-at-risk (OARs; bladder, rectum, sigmoid, small bowel) considered routinely in MRI-planned brachytherapy for locally advanced cervical cancer (LACC-BT), with emphasis on accuracy and efficiency, potentially practice changing.
Methods: Manually contoured OARs (Ground Truth; GT) on 200 T2-weighted 3D-MRIs from LACC-BT cases, treated with various applicators, were split for training|validation|testing (140|30|30). After initial investigation, nnU-Net was established as the best-performing network. The modified DL-model was iteratively tuned to identify ideal hyperparameters and performance metrics. Quantitative (Dice coefficient (DC), Hausdorff Distance 95th percentile (HD95), mean surface distance, precision, and Bland-Altman tests) and qualitative (5-point Likert scale grading by radiation oncologists; ROs) analyses evaluated geometric discrepancies relative to GT and clinical acceptance of predicted contours, respectively.
Results: The DL-model, generated following hyperparameter tuning and 5-fold cross validation, produced median DC|HD95(mm) for bladder of 0.94|2.00, rectum 0.88|3.80, sigmoid 0.77|13.23, and small bowel 0.66|21.04; lower metrics for OARs featuring larger GT volume variability. Bland-Altman analysis showed high degree of reliability (p>0.1) in GT vs DL-predicted volumes (95% within limits of agreement; mean difference across all OARs of <4.0 cc). ROs classified only 14% of predicted OARs as requiring major edits; remainder needed no or minor changes. DL-predicted contours, converted to DICOM, can easily be deployed for clinical use.
Conclusion: With high predictive accuracy and clinician acceptability, this modified DL-model can alleviate inter- and intra-operator OAR contouring variabilities and greatly improve complex, time-consuming workflows in MRI-based LACC-BT.
Scientific Session YIS – Presntation 3
RadiSeq: A whole-genome DNA sequencing simulation pipeline to investigate radiation-induced carcinogenesis
Felix Mathew, John Kildea
McGill University
Purpose: Behjati et al. (2016) reported evidence that radiotherapy-associated second tumours have a unique signature of radiation at the genomic level. However, our understanding of the mechanisms of radiation-induced carcinogenesis underlying these signatures is still limited. Single-cell DNA sequencing holds significant promise for probing the biophysical mechanisms underlying radiation carcinogenesis. However, experimental implementation is hampered by cost and time constraints. To address this, we present a novel computational pipeline for simulating radiation exposure and subsequent whole-genome DNA sequencing of human cells.
Methodology: Utilizing the TOPAS-nBIO Monte Carlo framework, we constructed a single-cell geometric nuclear DNA model for radiation-exposure simulations. Subsequently, we developed "RadiSeq", an application written in C++, to enable bulk-cell and single-cell whole-genome DNA sequencing simulations of radiation-exposed cell models from first principles. Users have the flexibility to specify parameters such as the sequencer model, sequencing coverage, number of cells to be sequenced, and sequencing protocol, among other parameters through a single input file. RadiSeq will then simulate the sequencing protocol, incorporating sequencing errors to ensure the specified sequencing quality and generate FASTQ files containing reads—data derived from small genomic segments analyzed during sequencing—similar to a real sequencing experiment.
Results: RadiSeq exhibits acceptable performance compared to real sequencing experiments, with early validation results showing promise.
Conclusion: It is imperative to expand our understanding of the biophysics of radiation-induced DNA damage to better characterize the carcinogenic risk of radiation. We believe that RadiSeq will be a useful tool for further investigations into the mechanisms of radiation carcinogenesis.
Scientific Session YIS – Presentation 4
Automating RT planning audits: benchmarking on a large cohort of prostate SABR patients
Conor Smith, Dominique Fortin
University of Victoria
Purpose: The evaluation of radiotherapy (RT) treatment plan quality is often performed on a patient basis or limited to small cohorts; long term trends are often left uncaptured. This study presents tools to perform automated audits on an RT database, and their application in assessing plan quality in prostate stereotactic ablative body radiotherapy (SABR).
Methods: By leveraging the Eclipse Scripting API’s ability to easily interface with the treatment database, and Python’s powerful data analysis tools, two custom scripts were developed to automatically extract and analyse various metrics and dose-volume histograms. The tools were benchmarked on three years of prostate SABR plans (n = 202) to evaluate the long-term effectiveness of RapidPlan (RP).
Results: Long term benefits of RP implementation have manifested in statistically significant reductions in mean and variance of intermediate rectum dose and femoral head max dose. However, the audit identified areas for RP model refinement as the mean and variance of the rectum V36Gy metric as well as bladder V36Gy and V33Gy variance slightly increased since the roll-out of RP. It was found that treatment planners rarely altered the initial optimization parameters, allowing RP to fully guide treatment optimization in 87% of the plans generated in the second year of RP model usage. Despite these challenges, all plans met clinical standards.
Conclusions: The automated tools were successfully benchmarked. Soon, they will be applied to audit the quality of treatment plans for many cancer sites and to evaluate the performance of future RP models before their clinical rollout.
Scientific Session YIS – Presentation 5
A step toward clinical translation: Developing a robust thermal calibration methodology for pyroelectric dosimeters.
Victoria Howard, James Renaud, Emily Heath, Bryan, Muir
Carleton University, National Research Council Canada,
Purpose: Accurate dose measurements for ultra-high dose rate (UHDR) therapy are challenging. This work presents a proposed solution based on pyroelectric calorimetry, where a determination of radiation induced temperature rise is made using current measured with readily-available clinical electrometers. The predicted dose sensitivity of different pyroelectric materials is investigated through thermal calibration.
Methods: The pyroelectric materials under investigation include lead zirconate titanate (PZT) crystals, lithium niobate (LiNbO_3) crystals and polyvinylidene fluoride (PVDF) films. Different holders were designed and constructed to establish electrical connections. Thermal calibration was done by placing the holder into a programmable water bath to vary the temperature while measuring current with an electrometer. The slope of the linear plot of current vs temperature rate, when divided by the area of the sample, gives the pyroelectric coefficient.
Results: The pyroelectric coefficients of the materials investigated were consistent with published values, although significant variability was observed in this work, as well as literature, demonstrating the importance of robust thermal calibrations. The expected detector response amounted to (4-6) pA (Gy/s)^-1 for PZT crystals, (45-48) pA (Gy/s)^-1 for LiNbO3 crystals, and (55-87) pA (Gy/s)^-1 for PVDF films. This means that the expected signal for the PVDF films in a UHDR beam will be on the order of 3500 pA.
Conclusions: Comparing the expected signal in a UHDR beam to the random variations observed during thermal calibrations, indicates that noise will contribute less than 0.2% to the reading, which demonstrates that pyroelectric calorimeters are promising dosimeters for UHDR beams.
Scientific Session YIS – Presentation 6
Using Colour Image Segmentation with Magnetic Resonance Images for Computed Tomography Synthesis.
Jules Faucher, Evan McNabb, Piotr Pater, Véronique Fortier, Ives R. Levesque
McGill University
Purpose: This work seeks a straightforward, fast, and deterministic synthetic computed tomography (sCT) method that correctly identifies soft tissue, air, and bone on magnetic resonance (MR) images using on well-establish colour-image segmentation algorithms, to be used in MR-based radiation dose calculation.
Methods: Three fast scans are acquired; an MPRAGE, a proton density-weighted (PDw) spiral ultra-short echo time (UTE) scan, and a second spiral UTE scan using the Ernst angle of bone for maximum bone signal. The sets of three images are used as colour channels in an RGB volume. Voxels are merged into supervoxels using the simple linear iterative clustering algorithm, which are then classified based on their colour. Air, fat and soft tissue voxels are assigned bulk CT numbers. Bone voxels are assigned CT numbers using the pixel values from PDw images. This method was tested in a phantom and in the head of a volunteer. The phantom featured segments of bovine femur, with and without marrow, and a Ping-Pong ball.
Results: For the most part, air and bone were well differentiated in both phantoms and in vivo, with exceptions arising in voxels that are mixtures of air and soft tissue. Some fat voxels were misclassified as bone. Minor geometrical distortions were observable in the spiral images, which propagated as inaccuracies in the sCT.
Conclusions: The algorithm can synthesize CT images of phantoms and head images, with other anatomical sites still requiring investigation. Fat suppression could be used to modify image contrast and improve fat segmentation.
Scientific Session YIS – Presentation 7
A deep learning pipeline for real-time conformal palliative radiotherapy of spine metastases
Ali-reza Haidari, Dal Granville, Elsayed Ali
Carleton University, Nova Scotia Health Authority, The Ottawa Hospital
Purpose: To develop a deep learning pipeline towards a CT simulation-free conformal palliative radiotherapy workflow for spinal metastases.
Methods: First, a dataset of planning CT (CTsim) and cone beam CT (CBCT) images from 220 patients, spanning the entire spine, was used to train and validate a novel two-stage generative adversarial network-based pipeline for synthetic CT (sCT) generation. Image quality was evaluated using a distinct dataset of 33 patients undergoing same-day treatment, with dosimetric analysis conducted on a subset. Next, a three-stage U-Net-based network was trained and validated for vertebral segmentation using an open-source dataset and was used to generate pseudo-ground truth vertebrae labeling on a local CTsim test dataset to evaluate performance on sCT images.
Results: The two-stage network significantly improved the Hounsfield Unit (HU) accuracy of CBCT images, reducing Mean Absolute Error from 225 ± 62 HU in CBCT to 86 ± 24 HU in sCT images, and the Mean Error was improved from 178 ± 91 HU to -8 ± 20 HU. Mean dose discrepancy was lowered by an average of 4.5%, with a 22% improvement in average gamma pass rate. The segmentation network achieved an average Dice Similarity Coefficient of 0.78 ± 0.12, Identification Rate of 93 ± 11 %, and Hausdorff Distance of 10.5 ± 4.4 mm, respectively.
Conclusions: By rapidly and accurately generating sCT from CBCT images, then automating individual vertebrae segmentation, this pipeline is a step towards real-time treatment planning and delivery of conformal palliative radiotherapy for spinal metastases within a single session.
Scientific Session YIS – Presentation 8
Point-scan Raman spectroscopy: A technique for high-spatial resolution radiation dosimetry.
Connor McNairn, Prarthana Pasricha, Wanye Gao, Andrea Payne, Edana Cassol, Vinita Chauhan, Sanjeena Dang, Jeffrey L. Andrews, Andrew Jirasek,Bryan Muir, Rowan M.Thomson, Sangeeta Murugkar
Carleton University, University of British Columbia
Purpose: To develop a highly sensitive micron-scale resolution dose readout system based on Raman spectroscopy (RS) and radiochromic films (RCF).
Methods: RCF samples (EBT-3) were irradiated using an X-ray source (6 MV Linac) to doses ranging between 0.3 – 2 Gy. The Raman spectra of RCFs were measured using a custom-built confocal Raman microscope with a spatial resolution of 1 µm at 785 nm excitation. Raman spectra were sampled in a 10x10 grid pattern over a 100 x 100 µm2 region of interest (ROI) on the film. Several spectra at each point were averaged to accurately estimate the Raman response at each pixel. These data underwent standard spectroscopic preprocessing, including a novel peak normalization technique.
Results: Several features in the EBT-3 Raman spectrum demonstrated significant response to ionizing radiation. The most prominent peak at 1445 cm-1 was utilized to evaluate the dose response of the average signal across the ROI of each sample. The 2260 cm-1 peak in the active layer is independent of dose and was used as a novel internal standard to normalize the entire Raman spectrum of the film. Raman response maps of the ROI demonstrated significant variance (8-12%) which is primarily attributed to heterogeneity in the film active layer. The average signal over the ROI of a scan varied between 1.6-2.2% for multiple measurements of the same film.
Conclusions: Our work demonstrates the excellent potential of point-scan Raman spectroscopy for high-spatial resolution radiation dosimetry due to its high sensitivity to low dose and 1 µm resolution.
Scientific Session YIS – Presentation 9
Tumor Nuclear Size as a Prognostic Biomarker in Post-radiotherapy Outcomes: Assessment of Pre-treatment Patient Histopathology and In-vitro Irradiation of HeLa Cells.
Yujing Zou, Behnaz Behmand, Manuela Pelmus, Michael D.C. Evans, Farhad Maleki, Magali Lecavalier-Barsoum, Shirin A. Enger
McGill University, University of Calgary, Jewish General Hospital
Nuclear DNA is the target of radiation-induced cell death or damage. A link may exist between tumor nuclear size, DNA content, and radiation-induced DNA double-strand breaks (DSBs), affecting patients’ clinical outcomes. However, this relationship is overlooked. This study investigates the impact of pre-treatment cell nuclear size variations on post-radiotherapy patient outcomes in gynecological malignancies and validates it in human cervical HeLa cancer cell lines. We examined nuclear size distributions from pre-treatment H&E-stained digital histopathology images of gynecological squamous cell carcinoma patients undergoing radiotherapy. A larger cancerous nuclear size distribution mean emerged as a significant prognostic factor associated with improved post-radiotherapy locoregional recurrence from multivariate Cox Proportional Hazard analysis (Hazard Ratio=0.36, p = 0.044), following the HPV status being the strongest indicator (p16 negative, Hazard Ratio=5.43, p=0.033). To explain this, 225 kV x-ray irradiation was delivered to the cervical HeLa cell line. Following irradiation, strongly positive Pearson’s correlations were observed between the number of γ-H2AX foci, serving as indicators of DSBs, and both the cell nuclear volume (r=0.84, 0.91, 0.88) and DNA content (r=0.77, 0.89, 0.83) for absorbed doses of 1 Gy, 2 Gy, and 4 Gy irradiation. Our findings establish a clear association between the greater number of radiation-induced DSBs resulting from larger cell nuclear volume and increased DNA content, which directly elucidates why gynecological patients with larger pre-treatment cancerous nuclear sizes tend to display more favorable post-radiotherapy outcomes. It emphasizes the role of pre-treatment cancerous nuclear size as a key prognostic indicator for post-radiotherapy locoregional recurrence for patients with gynecological cancers. Considering nuclear size variations for personalized dose prescriptions is essential for optimizing treatment outcomes in clinical practice.