An hetero-modal deep learning framework for medical image synthesis applied to contrast and non-contrast MRI.

Some pathologies such as cancer and dementia require multiple imaging modalities to fully diagnose and assess the extent of the disease. Magnetic resonance imaging offers this kind of polyvalence, but examinations take time and can require contrast agent injection. The flexible synthesis of these imaging sequences based on the available ones for a given patient could help reduce scan times or circumvent the need for contrast agent injection. In this work, we propose a deep learning architecture that can perform the synthesis of all missing imaging sequences from any subset of available images. The network is trained adversarially, with the generator consisting of parallel 3D U-Net encoders and decoders that optimally combines their multi-resolution representations with a fusion operation learned by an attention network trained conjointly with the generator network. We compare our synthesis performance with 3D networks using other types of fusion and a comparable number of trainable parameters, such as the mean/variance fusion. In all synthesis scenarios except one, the synthesis performance of the network using attention-guided fusion was better than the other fusion schemes. We also inspect the encoded representations and the attention network outputs to gain insights into the synthesis process, and uncover desirable behaviors such as prioritization of specific modalities, flexible construction of the representation when important modalities are missing, and modalities being selected in regions where they carry sequence-specific information. This work suggests that a better construction of the latent representation space in hetero-modal networks can be achieved by using an attention network.

External validation of the Memorial Sloan Kettering Cancer Center preoperative nomogram predicting lymph node invasion in a cohort of high-grade prostate cancer patients.

BACKGROUND: Commonly used preoperative nomograms predicting clinical and pathological outcomes in prostate cancer (PCa) patients have not been yet validated in high-grade only PCa patients. Our objective is to perform an external validation of the Memorial Sloan Kettering Cancer Center (MSKCC) preoperative nomogram as a predictor of lymph node invasion (LNI) in a cohort of high-grade PCa patients. METHODS: We included patients with high-grade PCa (Gleason ≥8) treated at our institution between 2011 and 2020 with radical prostatectomy and pelvic lymph node dissection without receiving neoadjuvant or adjuvant therapy. The area under the curve (AUC) of the receiver operator characteristic (ROC) was used to quantify the accuracy of the model to predict LNI. A calibration plot was used to evaluate the model's precision, and a decision curve analysis was computed to evaluate the net benefit associated with its use. This study was approved by our institution's ethics board. RESULTS: A total of 242 patients with a median age of 66 (60-71) years were included. LNI was observed in 70 (29%) patients with a mean of 16 (median = 15; range = 2-42) resected nodes. The MSKCC nomogram discriminative accuracy, as evaluated by the AUC-ROC was 79.0% (CI: [0.727-0.853]). CONCLUSION: The MSKCC preoperative nomogram is a good predictor of LNI and a useful tool associated with net clinical benefit in this patient population.

Unsupervised MRI motion artifact disentanglement: introducing MAUDGAN.

Objective.This study developed an unsupervised motion artifact reduction method for magnetic resonance imaging (MRI) images of patients with brain tumors. The proposed novel design uses multi-parametric multicenter contrast-enhanced T1W (ceT1W) and T2-FLAIR MRI images.Approach.The proposed framework included two generators, two discriminators, and two feature extractor networks. A 3-fold cross-validation was used to train and fine-tune the hyperparameters of the proposed model using 230 brain MRI images with tumors, which were then tested on 148 patients'in-vivodatasets. An ablation was performed to evaluate the model's compartments. Our model was compared with Pix2pix and CycleGAN. Six evaluation metrics were reported, including normalized mean squared error (NMSE), structural similarity index (SSIM), multi-scale-SSIM (MS-SSIM), peak signal-to-noise ratio (PSNR), visual information fidelity (VIF), and multi-scale gradient magnitude similarity deviation (MS-GMSD). Artifact reduction and consistency of tumor regions, image contrast, and sharpness were evaluated by three evaluators using Likert scales and compared with ANOVA and Tukey's HSD tests.Main results.On average, our method outperforms comparative models to remove heavy motion artifacts with the lowest NMSE (18.34±5.07%) and MS-GMSD (0.07 ± 0.03) for heavy motion artifact level. Additionally, our method creates motion-free images with the highest SSIM (0.93 ± 0.04), PSNR (30.63 ± 4.96), and VIF (0.45 ± 0.05) values, along with comparable MS-SSIM (0.96 ± 0.31). Similarly, our method outperformed comparative models in removingin-vivomotion artifacts for different distortion levels except for MS- SSIM and VIF, which have comparable performance with CycleGAN. Moreover, our method had a consistent performance for different artifact levels. For the heavy level of motion artifacts, our method got the highest Likert scores of 2.82 ± 0.52, 1.88 ± 0.71, and 1.02 ± 0.14 (p-values≪0.0001) for our method, CycleGAN, and Pix2pix respectively. Similar trends were also found for other motion artifact levels.Significance.Our proposed unsupervised method was demonstrated to reduce motion artifacts from the ceT1W brain images under a multi-parametric framework.

Commissioning and implementing a Quality Assurance program for dedicated radiation oncology MRI scanners.

PURPOSE: ACR and AAPM task group's guidelines addressing commissioning for dedicated MR simulators were recently published. The goal of the current paper is to present the authors' 2-year experience regarding the commissioning and introduction of a QA program based on these guidelines and an associated automated workflow. METHODS: All mandatory commissioning tests suggested by AAPM report 284 were performed and results are reported for two MRI scanners (MAGNETOM Sola and Aera). Visual inspection, vendor clinical or service platform, third-party software, or in-house python-based code were used. Automated QA and data analysis was performed via vendor, in-house or third-party software. QATrack+ was used for QA data logging and storage. 3D geometric distortion, B0 inhomogeneity, EPI, and parallel imaging performance were evaluated. RESULTS: Contrasting with AAPM report 284 recommendations, homogeneity and RF tests were performed monthly. The QA program allowed us to detect major failures over time (shimming, gradient calibration and RF interference). Automated QA, data analysis, and logging allowed fast ACR analysis daily and monthly QA to be performed in 3 h. On the Sola, the average distortion is 1 mm for imaging radii of 250 mm or less. For radii of up to 200 mm, the maximum, average (standard deviation) distortion is 1.2  and 0.4 mm (0.3 mm). Aera values are roughly double the Sola for radii up to 200 mm. EPI geometric distortion, ghosting ratio, and long-term stability were found to be under the maximum recommended values. Parallel imaging SNR ratio was stable and close to the theoretical value (ideal g-factor). No major failures were detected during commissioning. CONCLUSION: An automated workflow and enhanced QA program allowed to automatically track machine and environmental changes over time and to detect periodic failures and errors that might otherwise have gone unnoticed. The Sola is more geometrically accurate, with a more homogenous B0 field than the Aera.

MRI motion artifact reduction using a conditional diffusion probabilistic model (MAR-CDPM).

BACKGROUND: High-resolution magnetic resonance imaging (MRI) with excellent soft-tissue contrast is a valuable tool utilized for diagnosis and prognosis. However, MRI sequences with long acquisition time are susceptible to motion artifacts, which can adversely affect the accuracy of post-processing algorithms. PURPOSE: This study proposes a novel retrospective motion correction method named "motion artifact reduction using conditional diffusion probabilistic model" (MAR-CDPM). The MAR-CDPM aimed to remove motion artifacts from multicenter three-dimensional contrast-enhanced T1 magnetization-prepared rapid acquisition gradient echo (3D ceT1 MPRAGE) brain dataset with different brain tumor types. MATERIALS AND METHODS: This study employed two publicly accessible MRI datasets: one containing 3D ceT1 MPRAGE and 2D T2-fluid attenuated inversion recovery (FLAIR) images from 230 patients with diverse brain tumors, and the other comprising 3D T1-weighted (T1W) MRI images of 148 healthy volunteers, which included real motion artifacts. The former was used to train and evaluate the model using the in silico data, and the latter was used to evaluate the model performance to remove real motion artifacts. A motion simulation was performed in k-space domain to generate an in silico dataset with minor, moderate, and heavy distortion levels. The diffusion process of the MAR-CDPM was then implemented in k-space to convert structure data into Gaussian noise by gradually increasing motion artifact levels. A conditional network with a Unet backbone was trained to reverse the diffusion process to convert the distorted images to structured data. The MAR-CDPM was trained in two scenarios: one conditioning on the time step t $t$ of the diffusion process, and the other conditioning on both t $t$ and T2-FLAIR images. The MAR-CDPM was quantitatively and qualitatively compared with supervised Unet, Unet conditioned on T2-FLAIR, CycleGAN, Pix2pix, and Pix2pix conditioned on T2-FLAIR models. To quantify the spatial distortions and the level of remaining motion artifacts after applying the models, quantitative metrics were reported including normalized mean squared error (NMSE), structural similarity index (SSIM), multiscale structural similarity index (MS-SSIM), peak signal-to-noise ratio (PSNR), visual information fidelity (VIF), and multiscale gradient magnitude similarity deviation (MS-GMSD). Tukey's Honestly Significant Difference multiple comparison test was employed to quantify the difference between the models where p-value  < 0.05 $ < 0.05$ was considered statistically significant. RESULTS: Qualitatively, MAR-CDPM outperformed these methods in preserving soft-tissue contrast and different brain regions. It also successfully preserved tumor boundaries for heavy motion artifacts, like the supervised method. Our MAR-CDPM recovered motion-free in silico images with the highest PSNR and VIF for all distortion levels where the differences were statistically significant (p-values < 0.05 $< 0.05$ ). In addition, our method conditioned on t and T2-FLAIR outperformed (p-values < 0.05 $< 0.05$ ) other methods to remove motion artifacts from the in silico dataset in terms of NMSE, MS-SSIM, SSIM, and MS-GMSD. Moreover, our method conditioned on only t outperformed generative models (p-values < 0.05 $< 0.05$ ) and had comparable performances compared with the supervised model (p-values > 0.05 $> 0.05$ ) to remove real motion artifacts. CONCLUSIONS: The MAR-CDPM could successfully remove motion artifacts from 3D ceT1 MPRAGE. It is particularly beneficial for elderly who may experience involuntary movements during high-resolution MRI imaging with long acquisition times.

MedFusionGAN: multimodal medical image fusion using an unsupervised deep generative adversarial network.

PURPOSE: This study proposed an end-to-end unsupervised medical fusion generative adversarial network, MedFusionGAN, to fuse computed tomography (CT) and high-resolution isotropic 3D T1-Gd Magnetic resonance imaging (MRI) image sequences to generate an image with CT bone structure and MRI soft tissue contrast to improve target delineation and to reduce the radiotherapy planning time. METHODS: We used a publicly available multicenter medical dataset (GLIS-RT, 230 patients) from the Cancer Imaging Archive. To improve the models generalization, we consider different imaging protocols and patients with various brain tumor types, including metastases. The proposed MedFusionGAN consisted of one generator network and one discriminator network trained in an adversarial scenario. Content, style, and L1 losses were used for training the generator to preserve the texture and structure information of the MRI and CT images. RESULTS: The MedFusionGAN successfully generates fused images with MRI soft-tissue and CT bone contrast. The results of the MedFusionGAN were quantitatively and qualitatively compared with seven traditional and eight deep learning (DL) state-of-the-art methods. Qualitatively, our method fused the source images with the highest spatial resolution without adding the image artifacts. We reported nine quantitative metrics to quantify the preservation of structural similarity, contrast, distortion level, and image edges in fused images. Our method outperformed both traditional and DL methods on six out of nine metrics. And it got the second performance rank for three and two quantitative metrics when compared with traditional and DL methods, respectively. To compare soft-tissue contrast, intensity profile along tumor and tumor contours of the fusion methods were evaluated. MedFusionGAN provides a more consistent, better intensity profile, and a better segmentation performance. CONCLUSIONS: The proposed end-to-end unsupervised method successfully fused MRI and CT images. The fused image could improve targets and OARs delineation, which is an important aspect of radiotherapy treatment planning.

Patient-specific geometrical distortion corrections of MRI images improve dosimetric planning accuracy of vestibular schwannoma treated with gamma knife stereotactic radiosurgery.

PURPOSE: To investigate the impact of MRI patient-specific geometrical distortion (PSD) on the quality of Gamma Knife stereotactic radiosurgery (GK-SRS) plans of the vestibular schwannoma (VS) tumors. METHODS AND MATERIALS: Three open access datasets including the MPI-Leipzig Mind-Brain-Body (318 patients), the slow event-related fMRI designs dataset (62 patients), and the VS dataset (242 patients) were used. We used first two datasets to train a 3D convolution network to predict the distortion map of third dataset that were then used to calculate and correct the PSD. GK-SRS plans of VS dataset were used to evaluate dose distribution of PSD-corrected MRI images. GK-SRS prescription dose of VS cases was 12 Gy. Geometric and dosimetric discrepancies were assessed between the dose distributions and contours before and after the PSD corrections. Geometry indices were center of the contours, Dice coefficient (DC), Hausdorff distance (HD), and dosimetric indices were D µ ${D_\mu }$ , D m a x ${D_{max}}$ , D m i n ${D_{min}}$ , and D 95 % ${D_{95{\mathrm{\% }}}}$ doses, target coverage (TC), Paddick's conformity index (PCI), Paddick's gradient index (GI), and homogeneity index (HI). RESULTS: Geometric distortions of about 1.2 mm were observed at the air-tissue interfaces at the air canal and nasal cavity borders. Average center of the targets was significantly distorted along the frequency encoding direction after the PSD-correction. Average DC and HD metrics were 0.90 and 2.13 mm. Average D µ ${D_\mu }$ , D 95 % , ${D_{95{\mathrm{\% ,}}}}$ and D m i n ${D_{min}}$ in Gy significantly increased after PSD correction from 16.85 to 17.25, 12.30 to 12.77, and from 8.98 to 9.92. D m a x ${D_{max}}$ did not significantly change after the correction. Average TC and PCI significantly increased from 0.97 to 0.98, and 0.94 to 0.96. Average GI decreased significantly from 2.24 to 2.15 after PSD correction. However, HI did not significantly change after the correction. CONCLUSION: The proposed method could predict and correct the PSD that indicates the importance of PSD correction before GK-SRS plans of the VS patients.

Shuffle-ResNet: Deep learning for predicting LGG IDH1 mutation from multicenter anatomical MRI sequences.

Background and Purpose.The world health organization recommended to incorporate gene information such as isocitrate dehydrogenase 1 (IDH1) mutation status to improve prognosis, diagnosis, and treatment of the central nervous system tumors. We proposed our Shuffle Residual Network (Shuffle-ResNet) to predict IDH1 gene mutation status of the low grade glioma (LGG) tumors from multicenter anatomical magnetic resonance imaging (MRI) sequences including T2-w, T2-FLAIR, T1-w, and T1-Gd.Methods and Materials.We used 105 patient's dataset available in The Cancer Genome Atlas LGG project where we split them into training and testing datasets. We implemented a random image patch extractor to leverage tumor heterogeneity where about half a million image patches were extracted. RGB dataset were created from image concatenation. We used random channel-shuffle layer in the ResNet architecture to improve the generalization, and, also, a 3-fold cross validation to generalize the network's performance. The early stopping algorithm and learning rate scheduler were employed to automatically halt the training.Results.The early stopping algorithm terminated the training after 131, 106, and 96 epochs in fold 1, 2, and 3. The accuracy and area under the curve (AUC) of the validation dataset were 81.29% (95% CI (79.87, 82.72)) and 0.96 (95% CI (0.92, 0.98)) when we concatenated T2-FLAIR, T1-Gd, and T2-w to produce an RGB dataset. The accuracy and AUC values of the test dataset were 85.7% and 0.943.Conclusions.Our Shuffle-ResNet could predict IDH1 gene mutation status using multicenter MRI. However, its clinical application requires more investigation.

Knowledge-based planning algorithm for lung SBRT with robust Bayesian stochastic frontier analysis and missing data management.

PURPOSE: A knowledge-based planning technique is developed based on Bayesian stochastic frontier analysis. A novel missing data management is applied in order to handle missing organs-at-risk and work with a complete dataset. METHODS: Geometric metrics are used to predict DVH metrics for lung SBRT with a retrospective database of 299 patients. In total, 16 DVH metrics were predicted for the main bronchus, heart, esophagus, spinal cord PRV, great vessels, and chest wall. The predictive model is tested on a test group of 50 patients. RESULTS: Mean difference between the observed and predicted values ranges between 1.5 ± 1.9 Gy and 4.9 ± 5.3 Gy for the spinal cord PRV D0.35cc and the main bronchus D0.035cc, respectively. CONCLUSIONS: The missing data model implanted in the predictive model is robust in the estimation of the parameters. Bayesian stochastic frontier analysis with missing data management can be used to predict DVH metrics for lung SBRT treatment planning.

Direct in-water radiation dose measurements using Cherenkov emission corrected signals from polarization imaging for a clinical radiotherapy application.

Cherenkov emission (CE) is a visible blueish light emitted in water mediums irradiated by most radiotherapy treatment beams. However, CE is produced anisotropically which currently imposes a geometrical constraint uncertainty for dose measurements. In this work, polarization imaging is proposed and described as a method enabling precise 2D dose measurements using CE. CE produced in a water tank is imaged from four polarization angles using a camera coupled to a rotating polarizer. Using Malus' law, the polarized component of CE is isolated and corrected with Monte Carlo calculated CE polar and azimuthal angular distributions. Projected dose measurements resulting from polarization-corrected CE are compared to equivalent radiochromic film measurements. Overall, agreement between polarized corrected CE signal and films measurements is found to be within 3%, for projected percent depth dose (PPDD) and profiles at the different tested energies ([Formula: see text]: 6 and [Formula: see text], e[Formula: see text]: 6 and 18[Formula: see text]). In comparison, raw Cherenkov emission presented deviations up 60% for electron beam PPDDs and 20% for photon beams PPDDs. Finally, a degree of linear polarization between 29% and 47% was measured for CE in comparison to [Formula: see text]% for scintillation. Hence, polarization imaging is found to be a promising and powerful method for improved radio-luminescent dose measurements with possible extensions to signal separation.

Accurate dose measurements using Cherenkov emission polarization imaging.

PURPOSE: Cherenkov radiation carries the potential of direct in-water dose measurements, but its precision is currently limited by a strong anisotropy. Taking advantage of polarization imaging, this work proposes a new approach for high-accuracy Cherenkov emission dose measurements. METHODS: Cherenkov radiation produced in a 15 × 15 × 20-cm3 water tank is imaged with a cooled charge-coupled device (CCD) camera from four polarizer transmission axes [0, 45, 90, 135°]. The water tank is positioned at the isocenter of a 5 × 5-cm2 , 6-, and 18-MV photon beam. Using Malus' law, the polarized portion of the signal is extracted. Corrections are applied to the polarized signal following azimuthal and polar Cherenkov emission angular distributions extracted from Monte Carlo simulations. Projected percent depth dose and beam profiles are measured and compared with the prediction from a treatment planning system (TPS). RESULTS: Corrected polarized signals on the central axis reduced deviations at depth (mean ± standard deviation) from 8% ± 5% to 0.8% ± 1% at 6 MV and 8% ± 7% to 1% ± 3% at 18 MV. For the profile measurement, differences between the corrected polarized signal and the TPS calculations are 1% ± 3% and 2% ± 3% on the central axis at 6 and 18 MV respectively. In these conditions, Cherenkov emission is shown to be partly polarized. CONCLUSIONS: This work proposes a novel polarization imaging approach enabling high-precision water-based dose measurements using the Cherenkov radiation. The method allows a correction of the Cherenkov emission anisotropy within 4% on the beam central axis and in depth.

Advantages of TRUS-based delineation for high-dose-rate prostate brachytherapy planning.

PURPOSE: To evaluate the variability of prostate contours delineated on computed tomography (CT) and transrectal ultrasound (TRUS). MATERIAL AND METHODS: A TRUS-based high-dose-rate (HDR) brachytherapy procedure was introduced in 2016 in our center. The first thirty patients were additionally imaged with CT immediately after the treatment. In 2018, four different radiation oncologists (ROs: 1, 2, 3, 4) contoured the prostate on both modalities. A volume comparison was performed between CT and TRUS imaging. Using prostate gold fiducial makers, a rigid registration between CT and TRUS was done in 20 of the 30 patients studied. Jaccard index (JI) was computed to evaluate the inter-observer volume delineation agreement. RESULTS: The ratio of TRUS/CT volumes was 0.82 (95% CI: 0.79-0.87%). The mean JI was 87% for CT and 92% for TRUS, when comparing all four ROs; CT and TRUS JIs were significantly different (p < 0.001). The mean JI for the prostate on CT was significantly more consistent (p < 0.001) when comparing RO1, 2, and 3 together (RO1-2, RO1-3, and RO2-3; mean = 89%) than when comparing RO4 (newest to clinical practice) to others (RO1-4, RO2-4, and RO3-4; mean = 85%). For TRUS planning, the mean JI was not significantly different (p > 0.05) when comparing all ROs. CONCLUSIONS: The inter-observer and intra-observer variability were statistically significantly smaller on TRUS compared to CT-based planning, despite varying ROs clinical experiences. The superior soft tissue contrast offered by TRUS obviates the effect of the ROs experience on prostate contour volumes and enables more reproducible prostate delineation.

On the proper use of structural similarity for the robust evaluation of medical image synthesis models.

PURPOSE: To propose good practices for using the structural similarity metric (SSIM) and reporting its value. SSIM is one of the most popular image quality metrics in use in the medical image synthesis community because of its alleged superiority over voxel-by-voxel measurements like the average error or the peak signal noise ratio (PSNR). It has seen massive adoption since its introduction, but its limitations are often overlooked. Notably, SSIM is designed to work on a strictly positive intensity scale, which is generally not the case in medical imaging. Common intensity scales such as the Houndsfield units (HU) contain negative numbers, and they can also be introduced by image normalization techniques such as the z-normalization. METHODS: We created a series of experiments to quantify the impact of negative values in the SSIM computation. Specifically, we trained a three-dimensional (3D) U-Net to synthesize T2-weighted MRI from T1-weighted MRI using the BRATS 2018 dataset. SSIM was computed on the synthetic images with a shifted dynamic range. Next, to evaluate the suitability of SSIM as a loss function on images with negative values, it was used as a loss function to synthesize z-normalized images. Finally, the difference between two-dimensional (2D) SSIM and 3D SSIM was investigated using multiple 2D U-Nets trained on different planes of the images. RESULTS: The impact of the misuse of the SSIM was quantified; it was established that it introduces a large downward bias in the computed SSIM. It also introduces a small random error that can change the relative ranking of models. The exact values for this bias and error depend on the quality and the intensity histogram of the synthetic images. Although small, the reported error is significant considering the small SSIM difference between state-of-the-art models. It was shown therefore that SSIM cannot be used as a loss function when images contain negative values due to major errors in the gradient calculation, resulting in under-performing models. 2D SSIM was also found to be overestimated in 2D image synthesis models when computed along the plane of synthesis, due to the discontinuities between slices that is typical of 2D synthesis methods. CONCLUSION: Various types of misuse of the SSIM were identified, and their impact was quantified. Based on the findings, this paper proposes good practices when using SSIM, such as reporting the average over the volume of the image containing tissue and appropriately defining the dynamic range.

On the use of machine learning methods for mPSD calibration in HDR brachytherapy.

We sought to evaluate the feasibility of using machine learning (ML) algorithms for multipoint plastic scintillator detector (mPSD) calibration in high-dose-rate (HDR) brachytherapy. Dose measurements were conducted under HDR brachytherapy conditions. The dosimetry system consisted of an optimized 1-mm-core mPSD and a compact assembly of photomultiplier tubes coupled with dichroic mirrors and filters. An 192Ir source was remotely controlled and sent to various positions in a homemade PMMA holder, ensuring 0.1-mm positional accuracy. Dose measurements covering a range of 0.5 to 12 cm of source displacement were carried out according to TG-43 U1 recommendations. Individual scintillator doses were decoupled using a linear regression model, a random forest estimator, and artificial neural network algorithms. The dose predicted by the TG-43U1 formalism was used as the reference for system calibration and ML algorithm training. The performance of the different algorithms was evaluated using different sample sizes and distances to the source for the mPSD system calibration. We found that the calibration conditions influenced the accuracy in predicting the measured dose. The decoupling methods' deviations from the expected TG-43 U1 dose generally remained below 20%. However, the dose prediction with the three algorithms was accurate to within 7% relative to the dose predicted by the TG-43 U1 formalism when measurements were performed in the same range of distances used for calibration. In such cases, the predictions with random forest exhibited minimal deviations (<2%). However, the performance random forest was compromised when the predictions were done beyond the range of distances used for calibration. Because the linear regression algorithm can extrapolate the data, the dose prediction by the linear regression was less influenced by the calibration conditions than random forest. The linear regression algorithm's behavior along the distances to the source was smoother than those for the random forest and neural network algorithms, but the observed deviations were more significant than those for the neural network and random forest algorithms. The number of available measurements for training purposes influenced the random forest and neural network models the most. Their accuracy tended to converge toward deviation values close to 1% from a number of dwell positions greater than 100. In performing HDR brachytherapy dose measurements with an optimized mPSD system, ML algorithms are good alternatives for precise dose reporting and treatment assessment during this kind of cancer treatment.

External validation of a hidden Markov model for gamma-based classification of anatomical changes in lung cancer patients using EPID dosimetry.

PURPOSE: To externally validate a hidden Markov model (HMM) for classifying gamma analysis results of in vivo electronic portal imaging device (EPID) measurements into different categories of anatomical change for lung cancer patients. Additionally, the relationship between HMM classification and deviations in dose-volume histogram (DVH) metrics was evaluated. METHODS: The HMM was developed at CHU de Québec (CHUQ), and trained on features extracted from gamma analysis maps of in vivo EPID measurements from 483 fractions (24 patients, treated with three-dimensional 3D-CRT or intensity modulated radiotherapy), using the EPID measurement of the first treatment fraction as reference. The model inputs were the average gamma value, standard deviation, and average value of the highest 1% of gamma values, all averaged over all beams in a fraction. The HMM classified each fraction into one of three categories: no anatomical change (Category 1), some anatomical change (no clinical action needed, Category 2) and severe anatomical change (clinical action needed, Category 3). The external validation dataset consisted of EPID measurements from 263 fractions of 30 patients treated at Maastro with volumetric modulated arc therapy (VMAT) or hybrid plans (containing both static beams and VMAT arcs). Gamma analysis features were extracted in the same way as in the CHUQ dataset, by using the EPID measurement of the first fraction as reference (γQ), and additionally by using an EPID dose prediction as reference (γM). For Maastro patients, cone beam computed tomography (CBCT) scans and image-guided radiotherapy (IGRT) classification of these images were available for each fraction. Contours were propagated from the planning CT to the CBCTs, and the dose was recalculated using a Monte Carlo dose engine. Dose-volume histogram metrics for targets and organs-at-risk (OARs: lungs, heart, mediastinum, spinal cord, brachial plexus) were extracted for each fraction, and compared to the planned dose. HMM classification of the external validation set was compared to threshold classification based on the average gamma value alone (a surrogate for clinical classification at CHUQ), IGRT classification as performed at Maastro, and differences in DVH metrics extracted from 3D dose recalculations on the CBCTs. RESULTS: The HMM achieved 65.4%/65.0% accuracy for γQ and γM, respectively, compared to average gamma threshold classification. When comparing HMM classification with IGRT classification, the overall accuracy was 29.7% for γQ and 23.2% for γM. Hence, HMM classification and IGRT classification of anatomical changes did not correspond. However, there is a trend towards higher deviations in DVH metrics with classification into higher categories by the HMM for large OARs (lungs, heart, mediastinum), but not for the targets and small OARs (spinal cord, brachial plexus). CONCLUSION: The external validation shows that transferring the HMM for anatomical change classification to a different center is challenging, but can still be valuable. The HMM trained at CHUQ cannot be used directly to classify anatomical changes in the Maastro data. However, it may be possible to use the model in a different capacity, as an indicator for changes in the 3D dose based on two-dimensional EPID measurements.

Dosimetric performance of a multipoint plastic scintillator dosimeter as a tool for real-time source tracking in high dose rate 192 Ir brachytherapy.

PURPOSE: This study aims to present the performance of a multipoint plastic scintillation detector (mPSD) as a tool for real-time dose measurements (covering three orders of magnitude in dose rate), source-position triangulation, and dwell time assessment in high dose rate (HDR) brachytherapy. METHODS: A previously characterized and optimized three-point sensor system was used for HDR brachytherapy measurements. The detector was composed of three scintillators: BCF60, BCF12, and BCF10. Scintillation light was transmitted through a single 1-mm-diameter clear optical fiber and read by a compact assembly of photomultiplier tubes (PMTs). Each component was numerically optimized to allow for signal deconvolution using a multispectral approach, taking care of the Cerenkov stem effect as well as extracting the dose from each scintillator. The PMTs were read simultaneously using a data acquisition board at a rate of 100 KHz and controlled with in-house software based on Python. An 192 I r source (Flexitron, Elekta-Brachy) was remotely controlled and sent to various positions in a in-house PMMA holder, ensuring 0.1 mm positional accuracy. Dose measurements covering a range of 10 cm of source movement were carried out according to TG-43 U1 recommendations. Water measurements were performed in order to: (a) characterize the system's response in terms of angular dependence; (b) obtain the relative contribution of positioning and measurement uncertainties to the total system uncertainty; (c) assess the system's temporal resolution; and (d) track the source position in real time. The triangulation principle was applied to report the source position in three-dimensional space. RESULTS: As expected, the positioning uncertainty dominated close to the source, whereas the measurement uncertainty dominated at larger distances. A maximum measurement uncertainty of 17 % was observed for the BCF60 scintillator at 10 cm from the source. Based on the uncertainty chain, the best compromises between positioning and measurement uncertainties were reached at 17.2, 17.4, and 17.5 mm for the BCF10, BCF12, and BCF60 scintillators, respectively, which also corresponded to the recommended optimal distances to the source for calibration purposes. The detector further exhibited no angular dependence. All dose values were found to be within 2% of the dose value at 90 ∘ . In the experiments performed for source-position determination, the system provided an average location with a standard deviation under 1.7 mm. The maximum observed differences between measured and expected values were 1.82 and 1.8 mm in the x- and z-directions, respectively. Deviations between the mPSD measurements and expected TG-43 values were below 5% in all the explored measurement conditions. With regard to dwell time measurement accuracy, the maximum deviation observed at all distances was 0.56 ± 0.25 s, with a weighted average of the three scintillators below 0.33 ± 0.37 s at all distances covered in this study. CONCLUSIONS: Real-time HDR brachytherapy measurements were performed with an optimized mPSD system. The performance of the system demonstrated that it could be used for simultaneous, in vivo, real-time reporting of dose, dwell time, and source position during HDR brachytherapy.

Tomographic-based 3D scintillation dosimetry using a three-view plenoptic imaging system.

PURPOSE: To demonstrate the feasibility of a three-plenoptic camera projection, scintillation-based dosimetry system for measuring three-dimensional (3D) dose distributions of static photon radiation fields. METHODS: Static x-ray photon beams were delivered to a cubic plastic scintillator volume embedded within acrylic blocks. For each beam, three orthogonal projections of the scintillating light emission were recorded using a multifocus plenoptic camera. Experimental 3D reconstructions of the light distribution were obtained using an iterative maximum likelihood-expectation maximization (ML-EM) algorithm. For this purpose, the elements of the system matrix representing the contribution of the scintillator volume voxels to the camera sensor pixels were calculated using optical design software. A reconstruction-specific correction was applied to light reconstructions to account for scintillating light imaged by the camera but not directly resulting from dose deposition. Cross beam profiles (CBPs) and percentage depth dose (PDD) curves were compared to treatment planning system data for square fields. Three-dimensional and 3D gamma analyses were performed for concave-shaped dose distributions and the Pearson correlation coefficient and reconstruction error were employed to assess the quality of the measured relative 3D dose distributions. RESULTS: A full and accurate model of the plenoptic camera-based scintillation dosimetry system was implemented using the light ray tracing capabilities of optical design software. With this model, light distributions were successfully reconstructed over a volume of 60 × 60 × 60 mm 3 at a resolution of 2 mm. For relative 3D measurements of square radiation fields of 2 × 2 cm 2 , 3 × 3 cm 2 and 4 × 4 cm 2 compared with treatment planning system reference distributions, the maximum root-mean-square error of the CBPs evaluated at two different depths was of 3.2%, 1.2%, and 1.1%, respectively; as for the corresponding linearly fitted PDDs of the square fields, the slopes of the reconstructed dose distributions overestimated those of the reference distributions by at most 0.2%/ cm. The 2D gamma passing rate with a criterion of 2%/2 mm for the concave-shaped photon field was of 61.6%, 66.1%, and 76.4% using one, two, and three plenoptic projections; the respective success rates become 77.1%, 87.5%, and 94.9% using a criterion of 3%/3 mm. The 3D correlation coefficient for the corresponding reconstructions was of 0.688, 0.905, and 0.976, respectively. CONCLUSIONS: Three-dimensional light distributions emitted from within a plastic scintillator volume were successfully recovered using optical design software to establish a complete tomographic model of a plenoptic camera-based prototype. The tomographic model can equivalently extend to dynamic dose delivery measurements, providing temporal resolution limited by the camera's exposure time. This feasibility study enables a simplified design-to-implementation process for volumetric scintillation dosimetry prototypes toward fully meeting the clinical needs of 3D dose measurements for static and dynamic delivery techniques.

Simulating imaging-based tomographic systems using optical design software for resolving 3D structures of translucent media.

Imaging-based tomography is emerging as the technique of choice for resolving 3D structures of translucent media, in particular for applications in external beam radiation therapy and combustion diagnostics. However, designing experimental prototypes is time-consuming and costly, and is carried out without the certainty of the imaging optics being optimal. In this paper, we present an optical-design-software-based method that enables end-to-end simulation imaging-based tomography systems. The method, developed using the real ray tracing features of Zemax OpticStudio, was validated in the context of 3D scintillation dosimetry, where multiple imaging systems are used to image the 3D light pattern emitted within an irradiated cubic plastic scintillator volume. The flexibility of the workflow enabled the assessment and comparison of the tomographic performance of standard and focused plenoptic cameras for the reconstruction of a clinical radiation dose distribution. The versatility of the proposed method offers the potential to ease the developmental and optimization process of imaging systems used in volumetric emission computed tomography applications.

Stochastic frontier analysis as knowledge-based model to improve sparing of organs-at-risk for VMAT-treated prostate cancer.

Stochastic frontier analysis (SFA) is used as a novel knowledge-based technique in order to develop a predictive model of dosimetric features from significant geometric parameters describing a patient morphology. 406 patients treated with VMAT for prostate cancer were analyzed retrospectively. Cases were divided into three prescription-based groups. Seven geometric parameters are extracted to characterize the relationship between the organs-at-risk (bladder and rectum) with the planning volume (PTV). In total, 37 dosimetric parameters are tested for these two OARs. SFA allows the determination of the minimum achievable dose to the OAR based on the geometric parameters. Stochastic frontiers are determined with a maximum likelihood estimation technique. The SFA model was tested using validation cohort (30 patients with prescribed dose between 60 and 70 Gy) where 77% (23 out of 30) of the predicted DVHs present a 5% or less dose deterioration for the bladder and rectum with the planned DVH. SFA can be used in EBRT planning as a predictive model based on anatomical features of previously treated plans.