Papers by Jean-Claude Nunes
Image and vision computing, Jun 1, 2024
International Journal of Radiation Oncology Biology Physics, Dec 1, 2019
Various methods have recently been developed to generate pseudo-CT images for magnetic resonance ... more Various methods have recently been developed to generate pseudo-CT images for magnetic resonance imagingebased prostate dose planning. Several generative adversarial networks and U-Net deep learning methods with different loss functions and parameters were investigated in this study. In comparison with the patch-based method, these methods appear particularly promising for clinical use, owing to their low image and dose Purpose: Deep learning methods (DLMs) have recently been proposed to generate pseudo-CT (pCT) for magnetic resonance imaging (MRI) based dose planning. This study aims to evaluate and compare DLMs (U-Net and generative adversarial network [GAN]) using various loss functions (L2, single-scale perceptual loss [PL], multiscale PL, weighted multiscale PL) and a patch-based method (PBM). Methods and Materials: Thirty-nine patients received a volumetric modulated arc therapy for prostate cancer (78 Gy). T 2 -weighted MRIs were acquired in addition to planning CTs. The pCTs were generated from the MRIs using 7 configurations: 4 GANs (L2, single-scale PL, multiscale PL, weighted multiscale PL), 2 U-Net (L2 and single-scale PL), and the PBM. The imaging endpoints were mean absolute error and mean error, in Hounsfield units, between the reference CT (CT ref ) and the pCT. Dose uncertainties were quantified as mean absolute differences between the dose volume histograms (DVHs) calculated from the CT ref and pCT obtained by each method. Three-dimensional gamma indexes were analyzed. Results: Considering the image uncertainties in the whole pelvis, GAN L2 and U-Net L2 showed the lowest mean absolute error ( 34.4 Hounsfield units). The mean errors were not different than 0 (P .05). The PBM provided the highest uncertainties. Very
Zenodo (CERN European Organization for Nuclear Research), Jan 25, 2018
Dose calculation from MRI is a topical issue. New treatment systems combining a linear accelerato... more Dose calculation from MRI is a topical issue. New treatment systems combining a linear accelerator with a MRI have been recently being developed. MRI has good soft tissue contrast without ionizing radiation exposure. However, unlike CT, MRI does not provide electron density information necessary for dose calculation. We propose in this paper a machine learning method to simulate a CT from a target MRI and co-registered CT-MRI training set. Ten prostate MR and CT images have been considered. Firstly, a reference image was randomly selected in the training set. A common space has been built thanks to affine registrations between the training set and the reference image. Multiscale image descriptors such as spatial information, gradients and texture features were extracted from MRI patches at different levels of a Gaussian pyramid and used as voxel-wise characteristics in the learning scheme. A Conditional Inference Random Forest (CIRF) modelled the relation between MRI descriptors and CT patches. For validation, test images were spatially normalized and the same descriptors were computed to generate a new pCT. Leave-one out experiments were performed. We obtained a MAE = 45.79 (pCT vs CT). Dose volume histograms inside PTV and organs at risk are in close agreement. The D98% was 0.45 % (inside PTV) and the 3D gamma pass rate (1mm, 1%) was 99,2%. Our method has better results than direct bulk assignment. And the results suggest that the method may be used for dose calculations in an MR based planning system.
HAL (Le Centre pour la Communication Scientifique Directe), 2018
International audienc
HAL (Le Centre pour la Communication Scientifique Directe), 2022
HAL (Le Centre pour la Communication Scientifique Directe), 2022
HAL (Le Centre pour la Communication Scientifique Directe), Apr 6, 2011
International audienc
As new radiotherapy treatment systems using MRI (rather than traditional CT) are being developed,... more As new radiotherapy treatment systems using MRI (rather than traditional CT) are being developed, the accurate calculation of dose maps from MR imaging has become an increasing concern. MRI provides good soft-tissue but, unlike CT, lacks the electron density information necessary for dose calculation. In this paper, we proposed a generative adversarial network (GAN) using a perceptual loss to generate pseudo-CTs for prostate MRI dose calculation. This network was evaluated and compared to a U-Net network, a patch-based (PBM) and an atlas-based methods (ABM). Influence of the perceptual loss was assessed by comparing this network to a GAN using a L2 loss. GANs and U-Nets are rather similar with slightly better results for GANs. The proposed GAN outperformed the PBM by 9% and the ABM by 13% in term of MAE in whole pelvis. This method could be used for online dose calculation in MRI-only radiotherapy.
Cancer Radiotherapie, Jul 1, 2020
In context of head-and-neck radiotherapy, this study aims to compare MR image quality according t... more In context of head-and-neck radiotherapy, this study aims to compare MR image quality according to diagnostic (DIAG) and radiotherapy (RT) setups; and to optimise an MRI-protocol (including 3D T 1 and T 2 -weighted sequences) for dose-planning (based on pseudo-CT generation). Materials and methods. -To compare DIAG and RT setups, signal-to-noise-ratio (SNR) and percentageimage-uniformity (PIU) were computed on T 1 images of phantoms and volunteers. Influence of the sample conductivity on SNR was quantified using homemade phantoms. To obtain reliable T 1 and T 2 images for RT-planning, an experimental design was performed on volunteers by using SNR, contrast-to-noise-ratio (CNR) and mean-opinion-score (MOS). Further, pseudo-CTs were generated from 8 patients T 2 images with a state-of-art deep-learning method. These pseudo-CTs were evaluated by mean-absolute-error (MAE) and mean-error (ME). Results. -SNR was higher for DIAG-setup compared to RT-setup (SNR-ratio = 1.3). A clear influence of the conductivity on SNR was observed. PIU was higher for DIAG-setup (38.8%) compared to RT-setup (33.5%). Regarding the protocol optimisation, SNR, CNR, and MOS were 20.6, 6.16, and 3.91 for the optimal T 1 sequence. For the optimal T 2 sequence, SNR, CNR and MOS were 25.6, 44.46 and 4.0. In the whole head-and-neck area, the mean MAE and ME of the pseudo-CTs were 82.8 and -3.9 HU. Conclusion. -We quantified the image quality decrease induces by using an RT-setup for head-and-neck radiotherapy. To compensate this decrease, an MRI protocol was optimised by using an experimental design. This protocol of 15 minutes provides accurate images which could be used for MRI-dose-planning in clinical practice.
Physica Medica, Nov 1, 2022
Lecture Notes in Computer Science, 2003
This study introduces a new approach based on Bidimensional Empirical Mode Decomposition (BEMD) t... more This study introduces a new approach based on Bidimensional Empirical Mode Decomposition (BEMD) to extract texture features at multiple scales or spatial frequencies. Moreover, it can resolve the intrawave frequency modulation provided the frequency modulation. This decomposition, obtained by the bidimensional sifting process, plays an important role in the characterization of regions in textured images. The sifting process is realized using morphological operators to analyze the spatial frequencies and thanks to radial basis functions (RBF) for surface interpolation. We modified the original sifting algorithm to permit a pseudo bandpass decomposition of images by inserting scale criterion. Its effectiveness is demonstrated on synthetic and natural textures. In particular, we show that many different elements in textures can be extracted through the bidimensional empirical mode decomposition, which is fully unsupervised.
HAL (Le Centre pour la Communication Scientifique Directe), 2021
HAL (Le Centre pour la Communication Scientifique Directe), 2021
Cancer Radiotherapie, Oct 1, 2021
Cancer/Radiothérapie, 2018
IRBM, 2015
Abstract The paper describes an inexact tree-matching algorithm to register non-isomorphic 3D cor... more Abstract The paper describes an inexact tree-matching algorithm to register non-isomorphic 3D coronary artery trees over time. This work is carried out in the frame of the determination of the optimal viewing angles on the C-arm acquisition system for coronary percutaneous procedure. The matching method is based on association graph and maximum clique. Different similarity measures are compared, which use tree characteristics and geometric features of vascular branches. In order to take into account the topology variation between 3D vascular trees and thus improve the performance of the algorithm, we propose to insert artificial nodes in the association graph. Results show that unmatched node rate significantly decreases with the insertion of artificial nodes.
Uploads
Papers by Jean-Claude Nunes