COMP Highlights - Lombardy/Umbria
le 7 juin 2024 from 9h00 CDT to 10h30 CDT
Scientific Session – Highlights
Friday, June 7, 2024, 9:00-10:30
Scientific Session Highlights – Presentation 1
Investigation of Ge-doped fibre detectors as audit dosimeters.
Malcolm McEwen, Fatma Issa
National Research Council Canada, College of Medical Sciences and Technology
Purpose: To determine the expected ultimate accuracy of Ge-doped fibre dosimeters for small-field dosimetric audits. Optical fibre dosimeters have shown promise due to their small size (diameter ~ 120 μm) and ease of read-out (using standard TLD reader technology).
Methods: Commercial optical fibre was obtained and cut into 5 mm pieces for irradiation. In the initial investigation, a Co-60 beam was used to determine the sensitivity, linearity and repeatability in the range 1 Gy to 50 Gy. In the second investigation, linear accelerator beams were used to determine the energy dependence in MV photon beams. The main focus of the project was to determine influence quantities with the intention of identifying the ultimate accuracy of Ge-dope fibre dosimeters for clinical dose measurement.
Result: Pre-filtering of the dosimeter set was required to ensure similar levels of the Ge dopant, which strongly influences the dosimeter response. Measurements in Co-60 showed very good linearity up to 50 Gy, but lower doses were impacted by background measurements and TLD reader stability. For doses > 8 Gy, the standard uncertainty for a set of 10 fibres was 3.0%. Results for MV photon beams were similar in terms of precision. The energy dependence was obtained with an uncertainty of 1.5 % and showed a significant (~ 10% variation) from Co-60 to 25 MV.
Conclusions: The measurements in various high-energy photon beams indicate that low-cost Ge-doped fibre dosimeters can be used to measure absorbed dose to water in clinical beams with an uncertainty better than 4%.
Scientific Session Highlights – Presentation 2
Deep learning-based autocontouring algorithm for non-invasive intrafractional tumor-tracked radiotherapy (nifteRT) on Linac-MR.
Gawon Han, Keith Wachowicz, Nawaid Usmani, Don Yee, Jordan Wong, Arun Elangovan, B. Gino Fallone, Jihyun Yun,
University of Alberta, BC Cancer
Purpose: To develop a neural network-based tumor autocontouring algorithm with implementation of patient-specific hyperparameter optimization (HPO), and to validate its contouring accuracy using in-vivo MR images of liver, prostate, and lung cancer patients.
Methods: 2D intrafractional MR images were acquired at 4 frames/s using 3T MRI from 11 liver, 24 prostate, and 12 lung cancer patients. For each patient, 130 dynamic images were divided into 30 training, 30 validation, and 70 testing images. A U-Net architecture was applied for autocontouring, and was further improved by implementing HPO using CMA-ES. Six hyperparameters were optimized for each patient, for which intrafractional images and experts’ manual contours were input into the algorithm to find the optimal set of hyperparameters. For evaluation, Dice’s coefficient (DC), centroid displacement (CD), and Hausdorff distance (HD) were computed between the manual and autocontours. The performance of the algorithm was benchmarked against two standardized autosegmentation methods: non-optimized U-Net and nnU-Net.
Results: The algorithm was able to perform patient-specific HPO. The mean(standard deviation) DC, CD, and HD of the 47 patients were 0.92(0.04), 1.35(1.03) mm and 3.63(2.17) mm, respectively. Compared to the two benchmarking autosegmentation methods, the developed algorithm achieved the best overall performance in terms of contouring accuracy and speed.
Conclusion: To perform nifteRT on Linac-MR, an autocontouring scheme has been developed by implementing HPO to a U-Net-based autocontouring algorithm, and its contouring performance was evaluated using in-vivo MR images of various cancer patients. The developed algorithm performs patient-specific HPO enabling accurate tumor delineation comparable to that of experts.
Scientific Session Highlights – Presentation 3
Computed Tomography Synthesis in the Presence of Metal Implants.
Ives R. Levesque, Shogo Ushio, Piotr Pater, Véronique Fortier, Jules Faucher
McGill University
Purpose: To develop a synthetic computed tomography (sCT) method, from multi-echo magnetic resonance imaging (MRI), with accurate segmentation of bone and titanium implants and to assess it through recalculation of radiotherapy plan doses.
Methods: This work improves on a published sCT approach that segments regions of soft tissue, whole head, and MRI signal voids, all on one series of three-dimensional multi-echo gradient echo MR images. The signal void regions are separated into bone, air, and titanium regions using quantitative susceptibility mapping and morphological operations. sCT numbers are assigned to each voxel within the segmentations, accounting for partial volume effects. In nine patients treated for brain cancer, sCT numbers were compared to X-ray CT numbers. Dose distributions were recalculated on sCT for these patients and compared to the original treatment plans using gamma analysis.
Results: Segmentation of soft tissue, cortical bone and diploë was qualitatively good, with successful identification of titanium implants. The mean absolute error between sCT and CT numbers averaged across all datasets was 97 ± 16 HU (overall) and 159 ± 25 HU (bone regions only). The mean gamma pass rate (1% / 1 mm criterion) for dose distributions recalculated on sCTs was 99.15%, with a minimum gamma pass rate of 95.35%.
Conclusion: The CT synthesis method proposed in this work improves segmentation of bone and titanium implants and has comparable sCT number accuracy to other approaches in the literature. High average gamma pass rates suggest good potential for accurate dose calculations.
Scientific Session Highlights – Presentation 4
Raman spectroscopy and deep learning for tumour radiation response characterization.
Alejandra Maria Fuentes, Kirsty Milligan, Mitchell Wiebe, Apurva Narayan, Julian J. Lum, Alexandre G. Brolo, Jeffrey L. Andrews, Andrew Jirasek
UBC Okanagan, Western University, The University of Victoria
Purpose: Tumour cells exhibit altered metabolism that influences the extent of radiotherapy response. Understanding the biomolecular changes driving tumour radioresistance may assist with developing personalized treatment strategies. Raman spectroscopy is an optical modality that can be applied to characterize biomolecular radiation response in tumour cells and tissues. Convolutional Neural Networks (CNN) facilitate Raman spectral analysis by performing automated feature extraction directly from data by building predictive models, achieving excellent classification performance. Herein, a CNN is proposed to classify tumour cell radiation response according to its radiosensitivity.
Methods: The CNN was trained to characterize Raman spectra from three different cell lines as radiosensitive (LNCaP) or radioresistant (H460, MCF7) over a range of single-fraction doses and post-irradiation timepoints. Furthermore, the Gradient-Weighted Class Activation Mapping (Grad-CAM) technique was applied to visualize the Raman spectral features influencing the CNN predictions. Grad-CAM computes the relative contribution of each wavenumber in an input spectrum to the model prediction and displays it as heatmaps highlighting the most critical Raman peaks to the CNN classification.
Results: The CNN classified the spectra with accuracy, sensitivity, and specificity exceeding 99.8%. The Grad-CAM heatmaps exhibited distinct patterns of critical Raman features for radiosensitive and radioresistant cells. Heatmaps of H460 and MCF7 cells displayed strong contributions from glycogen and amino acids peaks. Conversely, LNCaP heatmaps exhibited contributions at lipid and phospholipid peaks.
Conclusion:The CNN/Grad-CAM framework can detect Raman spectral differences among tumour cells of varying radiosensitivity with high accuracy and identify potential signatures of radioresponse, facilitating automated tumour radiation response characterization.
Scientific Session Highlights – Presentation 5
HyperSight-ARCHER: Design of a clinical trial to investigate CBCT-based adaptive radiotherapy for head and neck cancer.
Amanda Cherpak, Robert Lee MacDonald, Derek Wilke,
QEII Cancer Centre, Dalhousie University
Purpose: To design a study investigating the use of HyperSight imaging solution’s cone beam CT (CBCT) and Ethos AI-informed planning software for adaptive for head and neck cancer treatment.
Methods: Study cohort will be 30 adult patients receiving radiation therapy for cancer of the head and neck, minimum of 21 fractions. HyperSight CBCTs will be taken on an Ethos unit after the treatment planning fan-beam CT (FBCT) and following fraction 21. Standard TrueBeam CBCTs (version 2.7 MR4) will also be collected throughout treatment. Objectives include i) comparison of image quality and dosimetry across imaging modalities, ii) plan adaptation and calculation directly on HyperSight CBCT, iii) evaluation of dosimetric improvements with adaptation, and iv) evaluation of a new metal artifact reduction (MAR) algorithm. Gains in workflow efficiencies will also be assessed.
Results: A protocol was finalized and underwent REB review. Investigational Testing Authorization was also acquired ahead of Health Canada licensing. Accrual began in August 2023 and should conclude by the end of 2024. Review of initial HyperSight data show image quality similar to FBCT and superior to TrueBeam 2.7 CBCT. Deformation of soft tissues can be seen throughout treatment, supporting the value of adaptive radiotherapy. The MAR algorithm decreased artifact from dental fillings, allowing for more accurate visualization of structures.
Conclusions: Initial results display superior performance of HyperSight CBCT compared to standard CBCT imaging. With full accrual, this trial will quantify the accuracy and efficiency of the new HyperSight imaging system, and its potential benefit to patients and clinical workflows.
Scientific Session Highlights – Presentation 6
Dosimetric Impacts of Automated Prostate Contouring in High Dose Rate Brachytherapy.
David DeVries, Noah Blackburn-Hum, Lucas Mendez, David D'Souza, Vikram Velker, Rohann Correa, Joelle Helou, Aaron Fenster, Doug Hoover
London Health Sciences Centre, University of Waterloo, Robarts Research Institute
Purpose: Automated contouring on ultrasound images during prostate cancer high dose rate brachytherapy would potentially reduce procedure time and improve contour consistency. This study characterizes the dosimetric impact of using automated contouring.
Methods: An AI model re-contoured the prostate on images from five treated brachytherapy patients. Five radiation oncologists (ROs) re-contoured each prostate four times generating two “manual” contours and two “AI-assisted” contours (AI contour as a starting point). For each contour, a new dose distribution/plan was developed using the same clinical needle positions. Fully automated contouring was investigated by overlaying dose distributions from the treatment plans created using AI contours onto the ROs’ manual contours to calculate V90, V100, V150 and V200. Baseline metrics were calculated by overlaying the dose from an RO’s manual contour onto the same RO’s second manual contour (intra-RO) or other ROs’ manual contours (inter-RO). AI-assisted contouring was investigated by overlaying the dose distributions developed using an RO’s AI-assisted contour onto either the same RO’s manual contours (intra-RO) or other ROs’ manual contours (inter-RO).
Results: Fully automated contouring demonstrated similar V90>99% and V100>95% pass rates (42%, 60%) compared to the intra-RO baseline (48%, 62%), and exceeded the inter-RO baseline (35%, 47%). AI-assisted contouring demonstrated enhanced passing rates for both intra-RO (53%, 63%) and inter-RO (50%, 59%). Median raw V90 and V100 values demonstrated a clear increase for the inter-RO case (p<0.001, p<0.001).
Conclusions: Using automated contouring during brachytherapy results in treatment plans that will acceptably treat the prostate, with AI-assisted contouring demonstrating the greatest potential.
Scientific Session Highlights – Presentation 7
Enhanced thermal neutron transportation modeling in TOPAS and cross validation with Geant4.
Sachin Dev, Felix Mathew, Professor John Kildea
McGill University
Purpose: In this study, we addressed the challenge of accurately modeling thermal neutron transportation in TOPAS by developing modular physics lists, coded in C++, to customize physics interactions. TOPAS relies on Geant4 as its underlying engine and the pre-made (reference) physics lists for thermal neutron transportation in TOPAS didn’t provide results comparable with Geant4. Our newly-developed TOPAS extension has modular physics lists and accurately handles thermal neutron transportation, utilizing TOPAS v3.6 which is a wrapper of Geant4 v10.06p03. Our results demonstrate comparable accuracy to Geant4, providing an efficient solution for thermal neutron transport within TOPAS for the first time.
Methods: To validate this TOPAS extension, a four-component soft tissue-equivalent sphere recommended by the International Commission on Radiation Units and Measurements (ICRU) was irradiated with monoenergetic neutron beams of 1 keV, 1 MeV, and 10 MeV. Secondary particle species spectra were recorded in three different scoring volumes placed at varying radial distances from the sphere’s centre and were compared with the spectra generated in the corresponding version of Geant4 for the identical simulation geometry.
Results: The subsequent recording and comparison of each secondary particle species (protons, electrons, deuterons, alphas, carbon nuclei, oxygen nuclei, nitrogen nuclei, and gamma) spectra between TOPAS and Geant4 yielded closely matched results across the diverse energy ranges, attesting the accuracy and reliability of the newly-developed TOPAS extension.
Conclusions: This development contributes to the capabilities of TOPAS, offering a validated and efficient solution for thermal neutron simulations in medical physics and beyond.