A convolutional neural network model for EPID-based non-transit dosimetry.

PURPOSE: To develop an alternative computational approach for EPID-based non-transit dosimetry using a convolutional neural network model. METHOD: A U-net followed by a non-trainable layer named True Dose Modulation recovering the spatialized information was developed. The model was trained on 186 Intensity-Modulated Radiation Therapy Step & Shot beams from 36 treatment plans of different tumor locations to convert grayscale portal images into planar absolute dose distributions. Input data were acquired from an amorphous-Silicon Electronic Portal Image Device and a 6 MV X-ray beam. Ground truths were computed from a conventional kernel-based dose algorithm. The model was trained by a two-step learning process and validated through a five-fold cross-validation procedure with sets of training and validation of 80% and 20%, respectively. A study regarding the dependance of the amount of training data was conducted. The performance of the model was evaluated from a quantitative analysis based the ϒ-index, absolute and relative errors computed between the inferred dose distributions and ground truths for six square and 29 clinical beams from seven treatment plans. These results were also compared to those of an existing portal image-to-dose conversion algorithm. RESULTS: For the clinical beams, averages of ϒ-index and ϒ-passing rate (2%-2mm > 10% Dmax ) of 0.24 (±0.04) and 99.29 (±0.70)% were obtained. For the same metrics and criteria, averages of 0.31 (±0.16) and 98.83 (±2.40)% were obtained with the six square beams. Overall, the developed model performed better than the existing analytical method. The study also showed that sufficient model accuracy can be achieved with the amount of training samples used. CONCLUSION: A deep learning-based model was developed to convert portal images into absolute dose distributions. The accuracy obtained shows that this method has great potential for EPID-based non-transit dosimetry.

Multimodal MRI-Based Whole-Brain Assessment in Patients In Anoxoischemic Coma by Using 3D Convolutional Neural Networks.

BACKGROUND: There is an unfulfilled need to find the best way to automatically capture, analyze, organize, and merge structural and functional brain magnetic resonance imaging (MRI) data to ultimately extract relevant signals that can assist the medical decision process at the bedside of patients in postanoxic coma. We aimed to develop and validate a deep learning model to leverage multimodal 3D MRI whole-brain times series for an early evaluation of brain damages related to anoxoischemic coma. METHODS: This proof-of-concept, prospective, cohort study was undertaken at the intensive care unit affiliated with the University Hospital (Toulouse, France), between March 2018 and May 2020. All patients were scanned in coma state at least 2 days (4 ± 2 days) after cardiac arrest. Over the same period, age-matched healthy volunteers were recruited and included. Brain MRI quantification encompassed both "functional data" from regions of interest (precuneus and posterior cingulate cortex) with whole-brain functional connectivity analysis and "structural data" (gray matter volume, T1-weighted, fractional anisotropy, and mean diffusivity). A specifically designed 3D convolutional neuronal network (CNN) was created to allow conscious state discrimination (coma vs. controls) by using raw MRI indices as the input. A voxel-wise visualization method based on the study of convolutional filters was applied to support CNN outcome. The Ethics Committee of the University Teaching Hospital of Toulouse, France (2018-A31) approved the study and informed consent was obtained from all participants. RESULTS: The final cohort consisted of 29 patients in postanoxic coma and 34 healthy volunteers. Coma patients were successfully discerned from controls by using 3D CNN in combination with different MR indices. The best accuracy was achieved by functional MRI data, in particular with resting-state functional MRI of the posterior cingulate cortex, with an accuracy of 0.96 (range 0.94-0.98) on the test set from 10-time repeated tenfold cross-validation. Even more satisfactory performances were achieved through the majority voting strategy, which was able to compensate for mistakes from single MR indices. Visualization maps allowed us to identify the most relevant regions for each MRI index, notably regions previously described as possibly being involved in consciousness emergence. Interestingly, a posteriori analysis of misclassified patients indicated that they may present some common functional MRI traits with controls, which suggests further favorable outcomes. CONCLUSIONS: A fully automated identification of clinically relevant signals from complex multimodal neuroimaging data is a major research topic that may bring a radical paradigm shift in the neuroprognostication of patients with severe brain injury. We report for the first time a successful discrimination between patients in postanoxic coma patients from people serving as controls by using 3D CNN whole-brain structural and functional MRI data. Clinical Trial Number http://ClinicalTrials.gov (No. NCT03482115).

Technical note: GAMMORA, a free, open-source, and validated GATE-based model for Monte-Carlo simulations of the Varian TrueBeam.

PURPOSE: Monte Carlo (MC) is the reference computation method for medical physics. In radiotherapy, MC computations are necessary for some issues (such as assessing figures of merit, double checks, and dose conversions). A tool based on GATE is proposed to easily create full MC simulations of the Varian TrueBeam STx. METHODS: GAMMORA is a package that contains photon phase spaces as a pre-trained generative adversarial network (GAN) and the TrueBeam's full geometry. It allows users to easily create MC simulations for simple or complex radiotherapy plans such as VMAT. To validate the model, the characteristics of generated photons are first compared to those provided by Varian (IAEA format). Simulated data are also compared to measurements in water and heterogeneous media. Simulations of 8 SBRT plans are compared to measurements (in a phantom). Two examples of applications (a second check and interplay effect assessment) are presented. RESULTS: The simulated photons generated by the GAN have the same characteristics (energy, position, and direction) as the IAEA data. Computed dose distributions of simple cases (in water) and complex plans delivered in a phantom are compared to measurements, and the Gamma index (3%/3mm) was always superior to 98%. The feasibility of both clinical applications is shown. CONCLUSIONS: This model is now shared as a free and open-source tool that generates radiotherapy MC simulations. It has been validated and used for five years. Several applications can be envisaged for research and clinical purposes.

A study of the interplay effect in radiation therapy using a Monte-Carlo model.

PURPOSE: In modulated radiotherapy, breathing motion can lead to Interplay (IE) and Blurring (BE) effects that can modify the delivered dose. The aim of this work is to present the implementation, the validation and the use of an open-source Monte-Carlo (MC) model that computes the delivered dose including these motion effects. METHODS: The MC model of the Varian TrueBeam was implemented using GATE. The dose delivered by different modulated plans is computed for several breathing patterns. A validation of these MC predictions is achieved by a comparison with measurements performed using a dedicated programmable motion platform, carrying a quality assurance phantom. A specific methodology was used to separate the IE and the BE. The influence of different motion parameters (period, amplitude, shape) and plan parameters (volume margin, dose per fraction) was also analyzed. RESULTS: The MC model was validated against measurement performed with motion with a mean 3D global gamma index pass rate of 97.5% (3%/3 mm). A significant correlation is found between the IE and the period and the antero-posterior amplitude of the motion but not between the IE and the CTV margin or the shape of motion. The results showed that the IE increases D2% and decreases the D98% of CTV with mean values of +6.9% and -3.3% respectively. CONCLUSIONS: We validated the feasibility to assess the IE using a MC model. We found that the most important parameter is the number of breathing cycles that must be greater than 20 for one arc to limit the IE.

Dosimetric performance of continuous EPID imaging in stereotactic treatment conditions.

PURPOSE: This study aims at investigating the dosimetric characteristics of a Varian aS1000 EPID, focusing on its continuous acquisition mode under the challenging conditions that can be met in stereotactic radiotherapy verification. METHODS: An aS1000 EPID installed on a Varian TrueBeamSTx was irradiated with 6 and 10 MV unflattened and flattened photon beams. In order to avoid detector saturation, the source-to-detector distance (SDD) was set to 150 or 180 cm depending on the dose rate. EPID image sets were acquired in continuous mode (CM) and also in the commonly used integrated mode (IM) for comparison, to evaluate dose linearity (including dose rate dependence), repeatability, reproducibility, stability, ghosting effect and field size dependence. RESULTS: CM response linearity was found to be within 0.8% of IM and independent of dose rate. Response repeatability was slightly better for IM and FF beams, being in all cases within 0.9%. Reproducibility was within 0.6% for both modes and all beam qualities. Response stability between continuous frames varied within 1% for dynamic and static irradiations and for all the beam qualities, showing its independence from these parameters. Ghosting effect was not significant, being comparable to signal variations between continuous frames (±1%). Field size dependence in both modes agreed within 1%. CONCLUSIONS: The dosimetric response of the aS1000 EPID in CM with FFF beams and high dose rates is comparable to that in IM and for flattened beams provided that the appropriate SDD is used. aS1000 EPID in continuous acquisition mode is therefore suitable for stereotactic applications.

A study of the interplay effect for VMAT SBRT using a four-axes motion phantom.

PURPOSE: To assess the accuracy of volumetric modulated arc therapy (VMAT) stereotactic body radiation therapy (SBRT) when treating moving targets (such as lung or liver lesions), focusing on the impact of the interplay effect in the event of complex breathing motion and when a gating window is used. METHODS: A dedicated programmable motion platform was implemented. This platform can carry large quality assurance (QA) phantoms and achieve complex three-dimensional (3D) motion. Volumetric modulated arc therapy SBRT plans were delivered with TrueBeam linac to this moving setup and the measured dose was compared to the computed one. Several parameters were assessed such as breathing period, dose rate, dose prescription, shape of the breathing pattern, the use of a planning target volume (PTV) margin, and the use of a gating window. RESULTS: Loss of dose coverage (D95%) was acceptable in most situations. The doses received by 95% of the CTV, D95% ( C T V m ) ranged from 94 to 101% (mean 98%) and the doses received by 2% of the CTV D2% ( C T V m ) ranged from 94% to 110% of the prescribed dose. A visible interplay effect was observed when no margin was used or when the number of breathing cycles during the treatment delivery was lower than 20. CONCLUSIONS: In our clinical context, treating lung and liver lesions using VMAT SBRT is reasonable. The interplay effect was moderated and acceptable in all simulated situations.

2D EPID dose calibration for pretreatment quality control of conformal and IMRT fields: A simple and fast convolution approach.

PURPOSE: This work presents an original algorithm that converts the signal of an electronic portal imaging device (EPID) into absorbed dose in water at the depth of maximum. METHODS: The model includes a first image pre-processing step that accounts for the non-uniformity of the detector response but also for the perturbation of the signal due to backscatter radiation. Secondly, the image is converted into absorbed dose to water through a linear conversion function associated with a dose redistribution kernel. These two computation parameters were modelled by correlating the on-axis EPID signal with absorbed dose measurements obtained on square fields by using an ionization chamber placed in water at the depth of maximum dose. The accuracy of the algorithm was assessed by comparing the dose determined from the EPID signal with the dose derived by the treatment planning system (TPS) using the ϒ-index. These comparisons were performed on 8 conformal radiotherapy treatment fields (3DCRT) and 18 modulated fields (IMRT). RESULTS: For a dose difference and a distance-to-agreement set to 3% of the maximum dose and 2 mm respectively, the mean percentage of points with a ϒ-value less than or equal to 1 was 99.8% ± 0.1% for 3DCRT fields and 96.8% ± 2.7% for IMRT fields. Moreover, the mean gamma values were always less than 0.5 whatever the treatment technique. CONCLUSION: These results confirm that our algorithm is an accurate and suitable tool for clinical use in a context of IMRT quality assurance programmes.

Nonlinearity in MCF7 Cell Survival Following Exposure to Modulated 6 MV Radiation Fields: Focus on the Dose Gradient Zone.

The study of cell survival following exposure to nonuniform radiation fields is taking on particular interest because of the increasing evidence of a nonlinear relationship at low doses. We conducted in vitro experiments using the MCF7 breast cancer cell line. A 2.4 × 2.4 cm(2) square area of a T25 flask was irradiated by a Varian Novalis accelerator delivering 6 MV photons. Cell survival inside the irradiation field, in the dose gradient zone and in the peripheral zone, was determined using a clonogenic assay for different radiation doses at the isocenter. Increased cell survival was observed inside the irradiation area for doses of 2, 10, and 20 Gy when nonirradiated cells were present at the periphery, while the cells at the periphery showed decreased survival compared to controls. Increased survival was also observed at the edge of the dose gradient zone for cells receiving 0.02 to 0.01 Gy when compared with cells at the periphery of the same flask, whatever the isocenter dose. These data are the first to report cell survival in the dose gradient zone. Radiotherapists must be aware of this nonlinearity in dose response.

Integration method of 3D MR spectroscopy into treatment planning system for glioblastoma IMRT dose painting with integrated simultaneous boost.

BACKGROUND: To integrate 3D MR spectroscopy imaging (MRSI) in the treatment planning system (TPS) for glioblastoma dose painting to guide simultaneous integrated boost (SIB) in intensity-modulated radiation therapy (IMRT). METHODS: For sixteen glioblastoma patients, we have simulated three types of dosimetry plans, one conventional plan of 60-Gy in 3D conformational radiotherapy (3D-CRT), one 60-Gy plan in IMRT and one 72-Gy plan in SIB-IMRT. All sixteen MRSI metabolic maps were integrated into TPS, using normalization with color-space conversion and threshold-based segmentation. The fusion between the metabolic maps and the planning CT scans were assessed. Dosimetry comparisons were performed between the different plans of 60-Gy 3D-CRT, 60-Gy IMRT and 72-Gy SIB-IMRT, the last plan was targeted on MRSI abnormalities and contrast enhancement (CE). RESULTS: Fusion assessment was performed for 160 transformations. It resulted in maximum differences <1.00 mm for translation parameters and ≤1.15° for rotation. Dosimetry plans of 72-Gy SIB-IMRT and 60-Gy IMRT showed a significantly decreased maximum dose to the brainstem (44.00 and 44.30 vs. 57.01 Gy) and decreased high dose-volumes to normal brain (19 and 20 vs. 23% and 7 and 7 vs. 12%) compared to 60-Gy 3D-CRT (p < 0.05). CONCLUSIONS: Delivering standard doses to conventional target and higher doses to new target volumes characterized by MRSI and CE is now possible and does not increase dose to organs at risk. MRSI and CE abnormalities are now integrated for glioblastoma SIB-IMRT, concomitant with temozolomide, in an ongoing multi-institutional phase-III clinical trial. Our method of MR spectroscopy maps integration to TPS is robust and reliable; integration to neuronavigation systems with this method could also improve glioblastoma resection or guide biopsies.

Cortical imaging on a head template: a simulation study using a resistor mesh model (RMM).

The T1 head template model used in Statistical Parametric Mapping Version 2000 (SPM2), was segmented into five layers (scalp, skull, CSF, grey and white matter) and implemented in 2 mm voxels. We designed a resistor mesh model (RMM), based on the finite volume method (FVM) to simulate the electrical properties of this head model along the three axes for each voxel. Then, we introduced four dipoles of high eccentricity (about 0.8) in this RMM, separately and simultaneously, to compute the potentials for two sets of conductivities. We used the direct cortical imaging technique (CIT) to recover the simulated dipoles, using 60 or 107 electrodes and with or without addition of Gaussian white noise (GWN). The use of realistic conductivities gave better CIT results than standard conductivities, lowering the blurring effect on scalp potentials and displaying more accurate position areas when CIT was applied to single dipoles. Simultaneous dipoles were less accurately localized, but good qualitative and stable quantitative results were obtained up to 5% noise level for 107 electrodes and up to 10% noise level for 60 electrodes, showing that a compromise must be found to optimize both the number of electrodes and the noise level. With the RMM defined in 2 mm voxels, the standard 128-electrode cap and 5% noise appears to be the upper limit providing reliable source positions when direct CIT is used. The admittance matrix defining the RMM is easy to modify so as to adapt to different conductivities. The next step will be the adaptation of individual real head T2 images to the RMM template and the introduction of anisotropy using diffusion imaging (DI).

Methylphenidate modulates cerebral post-stroke reorganization.

OBJECTIVE: We hypothesized that a single dose of methylphenidate (MP) would modulate cerebral motor activation and behavior in patients having suffered a subcortical stroke. METHODS: Eight men with a single stroke on the corticospinal tract resulting in a pure motor hemiparesia were included in a randomized, cross-over, double-blind, placebo-controlled study. Patients were first evaluated 17 days after stroke onset by validated neurological scales, motor tests and fMRI (flexion/extension of the digits) after 20 mg MP or placebo. Seven days later, the patients underwent the same protocol and received the drug they had not taken at the first evaluation. Each patient was his own control. RESULTS: Placebo intake did not change performance. MP compared to placebo elicited a significant improvement in motor performance of the affected hand at the finger tapping test. MP induced: (1) a hyperactivation of the ipsilesional primary sensorimotor cortex including the motor hand and face areas and of the contralesional premotor cortex; (2) a hypoactivation of the ipsilesional anterior cingulum. Hyperactivation in the face motor area correlated positively with the improvement in performance. CONCLUSION: We demonstrated that the reorganized network may efficiently be targeted by the drug and that the effect of MP might partly rely on an improvement in attention/effort through cingulum modulation.

Modulation of behavior and cortical motor activity in healthy subjects by a chronic administration of a serotonin enhancer.

UNLABELLED: SSRIs are postulated to modulate motor behavior. A single dose of selective serotoninergic reuptake inhibitors (SSRIs) like fluoxetine, paroxetine, or fluvoxamine, has been shown to improve motor performance and efficiency of information processing for simple sensorimotor tasks in healthy subjects. At a cortical level, a single dose of SSRI was shown to induce a hyperactivation of the primary sensorimotor cortex (S1M1) involved in the movement (Loubinoux, I., Boulanouar, K., Ranjeva, J. P., Carel, C., Berry, I., Rascol, O., Celsis, P., and Chollet, F., 1999. Cerebral functional magnetic resonance imaging activation modulated by a single dose of the monoamine neurotransmission enhancers fluoxetine and fenozolone during hand sensorimotor tasks. J. Cereb. Blood Flow Metab. 19 1365--1375, Loubinoux, I., Pariente, J., Boulanouar, K., Carel, C., Manelfe, C., Rascol, O., Celsis, P., and Chollet, F., 2002. A Single Dose of Serotonin Neurotransmission Agonist Paroxetine Enhances Motor Output. A double-blind, placebo-controlled, fMRI study in healthy subjects. NeuroImage 15 26--36). Since SSRIs are usually given for several weeks, we assessed the behavioral and cerebral effects of a one-month chronic administration of paroxetine on a larger group. In a double-blind, placebo controlled and crossover study, 19 subjects received daily 20 mg paroxetine or placebo, respectively, over a period of 30 days separated by a wash-out period of 3 months. After each period, the subjects underwent an fMRI (active or passive movement, dexterity task, sensory discrimination task) and a behavioral evaluation. Concurrently, a TMS (transcranial magnetic stimulation) study was conducted (Gerdelat-Mas, A., Loubinoux, I., Tombari, D., Rascol, O., Chollet, F., Simonetta-Moreau, M., 2005. Chronic administration of selective serotonin re-uptake inhibitor (SSRI) paroxetine modulates human motor cortex excitability in healthy subjects. NeuroImage 27,314--322). RESULTS: On the one hand, paroxetine improved motor performances at the finger tapping test (P=0.02) without affecting choice reaction time, strength and dexterity significantly. Subjects were also faster in processing the spatial incongruency between a stimulus and the motor response (P=0.04). In order to differentiate behavioral components, a principal component analysis was performed on all motor tests, and several characteristics were differentiated: strength, speed, skill, attention, and motor response coding. Paroxetine would improve the efficiency of motor response coding (MANOVA on the factors; factor 3, P=0.01). On the other hand, the chronic administration induced a significant hypoactivation of S1M1 whatever the task: motor or sensory, simple or complex (random effect analysis, P<0.05). The hypoactivation correlated with the improvement of performances at the finger tapping test (P<0.05) suggesting more efficiency in cerebral motor processing. CONCLUSIONS: Our results showed a clear modulation of sensory and motor cerebral activation after a chronic paroxetine administration. An improvement in both behavior and cerebral efficiency was suggested. It could be hypothesized that monoamines, by an unspecific effect, may tune the response of pyramidal neurons to optimize performances.

Effects of skull thickness, anisotropy, and inhomogeneity on forward EEG/ERP computations using a spherical three-dimensional resistor mesh model.

Bone thickness, anisotropy, and inhomogeneity have been reported to induce important variations in electroencephalogram (EEG) scalp potentials. To study this effect, we used an original three-dimensional (3-D) resistor mesh model described in spherical coordinates, consisting of 67,464 elements and 22,105 nodes arranged in 36 different concentric layers. After validation of the model by comparison with the analytic solution, potential variations induced by geometric and electrical skull modifications were investigated at the surface in the dipole plane and along the dipole axis, for several eccentricities and bone thicknesses. The resistor mesh permits one to obtain various configurations, as local modifications are introduced very easily. This has allowed several head models to be designed to study the effects of skull properties (thickness, anisotropy, and heterogeneity) on scalp surface potentials. Results show a decrease of potentials in bone, depending on bone thickness, and a very small decrease through the scalp layer. Nevertheless, similar scalp potentials can be obtained using either a thick scalp layer and a thin skull layer, and vice versa. It is thus important to take into account skull and scalp thicknesses, because the drop of potential in bone depends on both. The use of three different layers for skull instead of one leads to small differences in potential values and patterns. In contrast, the introduction of a hole in the skull highly increases the maximum potential value (by a factor of 11.5 in our case), because of the absence of potential drop in the corresponding volume. The inverse solution without any a priori knowledge indicates that the model with the hole gives the largest errors in both position and dipolar moment. Our results indicate that the resistor mesh model can be used as a robust and user-friendly simulation tool in EEG or event-related potentials. It makes it possible to build up real head models directly from anatomic magnetic resonance imaging without tessellation, and is able to take into account head heterogeneities very simply by changing volume elements conductivity. Hum. Brain Mapping 21:84-95, 2004.