Loading…

Flash Player 9 (or above) is needed to view presentations.
We have detected that you do not have it on your computer. To install it, go here.

Like this presentation? Why not share!

Simon David

on

  • 298 views

 

Statistics

Views

Total Views
298
Views on SlideShare
298
Embed Views
0

Actions

Likes
0
Downloads
1
Comments
0

0 Embeds 0

No embeds

Accessibility

Upload Details

Uploaded via as Microsoft PowerPoint

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment
  • Thank you Mr. the chairman and good afternoon every body. I’m here to talk about the application to the multi-observation fusion to clinical application such as multi-tracer image fusion and patient following.
  • Here is the outline of my talk. I’ll first present the context of oncology with the description of the of the positron emission tomography imaging and the external radiotherapy. Then I’ll emphasize the use of PET in clinical application with the patient monitoring and the multi-tracer analysis, which will lead me to the presentation of the multi-observation fusion. And finally I’ll conclude with our preliminary results and the further work.
  • First of all, in 2002, 11 millions of new cases of cancer were detected with 7 millions deaths. The prediction in 2030 are 11 millions death. On the one hand, the diagnosis of cancer is provided by different images modalities such as computed tomography, the magnetic resonance imaging and the emission imaging. Since 2000, PET / CT are combined. On the other hand, there is 3 main modalities in the cancer treatment which are the surgery, the chemotherapy and the radiotherapy. Most of the time, theses there modalities are combined. In my presentation, I’ll focus on the PET imaging and its potential application in radiotherapy and therapy response assessment.
  • The positron emission tomography the an functional imaging, will allow the visualization of physiological processes. The PET imaging is mainly used for cancer diagnosis and tumor staging. We inject a radiotracer composed by a biological tracer, which will target the tumor and a radionuclide, which is a beta emitter. The disintegration of the radionuclide will lead to the emission of 2 gamma rays, which will be detected by the scanner. Since 2000, there is new scanner generation where the PET and CT are acquired in the same bed. The PET/CT imaging is became the gold standard for the diagnosis.
  • The positron emission tomography the an functional imaging, will allow the visualization of physiological processes. The PET imaging is mainly used for cancer diagnosis and tumor staging. We inject a radiotracer composed by a biological tracer, which will target the tumor and a radionuclide, which is a beta emitter. The disintegration of the radionuclide will lead to the emission of 2 gamma rays, which will be detected by the scanner. Since 2000, there is new scanner generation where the PET and CT are acquired in the same bed. The PET/CT imaging is became the gold standard for the diagnosis.
  • And now of radiotherapy. The principle of radiotherapy is to use the ionizing radiation to kill the malignant cells. There is 3 kind of radiotherapy and the mainly used is the external radiotherapy where the photon or electron beams are produced by a linear accelerator. The shape of the beam is defined by the collimator. A crucial step in radiotherapy is the definition of the target volumes. 3 volumes are defined : - the gross tumor volume, defined with the conventional imaging modalities - the target volume, computed considering the anatomical information - and the planning target volume, take into account of the physiological variation of the patient.
  • Since few years, there is a development of the intensity modulated radiation therapy which make possible the administration of a non-uniform delivered dose. Therefore, we can adapt the treatment to the patient and reduce the irradiation of the organ at risk (OAR) and surrounding healthy tissues. In the following pictures, we can see the target volume definition effected by the PET/CT scan and the corresponding IMRT planning. Our aim is to improve the tumor volume definition through the analysis of multi-tracer scans. The 18 F-FDG is the radiotracer, the most widely used in PET, measure the glucose consumption. The tumor, being highly glucose consuming, will attracted the radiotracer. However, the FDG is not tumor specific because other physiological processes needs glucoses. Moreover, FDG target also inflammatory tissues leading to a miss-interpretation of the scan.
  • Recently, there is development of specific radio-marker measuring different tumor features. For example, the FMISO other similar-tracer measure the tumor hypoxia, meaning lack of oxygen. The hypoxia induce resilience to the radiotherapy, that’s why hypoxic tumor require a dose boosting. Another interesting radiotracer is the FLT, measuring the tumor proliferation. FLT, targeting the DNA , is a tumor specific radio-maker and avoid the uptake in inflammatory tissues contrary to the FDG. Yet, the FLT uptake in the tumor is generally lower than the FDG. Finally, we aimed to merging all the features obtain with the different radio-tracers. Each radiotracer measuring a specific biologic process, the scan not quantitatively comparable.
  • Another clinical application with PET is the patient follow-up. For patient, who underwent chemo or radiotherapy, it’s mandatory to early assess the response to the treatment. An estimation of the prognosis of the computable by analyzing PET scan acquired at different time of the treatment. An early assessment of the response to the treatment allow to adapt the therapy to the response and avoid toxic and costly cares for non-responding patients.
  • So, we are in a framework of multi-sensor observation of an object, with many clinical data available with the multi-modality imaging. We aimed to fuse all the available measurement obtained with the different radiotracer, in order to improve the tumor volume definition and at different time of the treatment to improve the assessment of the tumor response. We have used the statistical segmentation framework in order to model the information of multi-tracer and follow-scans. We were inspired from method develop in astronomical framework.
  • The method, developed by Flitti et al. describe a multi-band segmentation of a spectroscopic line data cube. The observation, obtained by an radio interferometers, are in 3 dimensions with the astronomical coordinates and the frequency on the third axis. The first step describe in the method is a reduction of the dimension problem. Considering the Gaussian mixture model, the expectation maximization algorithm fitted all the spectrum and assess mean and variance associate to each class. Then, the 6 most pertinent Gaussian where selected by using the K-mean algorithm. The next step consisted in computation of weight associated to each class with the Levenberg-Marquardt algorthim.
  • Dealing with reduce dimension, the line data cube are the classified using the Bayesian inference. Flitti et al. used Hierarchical Markovian Model, which take in account the spatial dependencies between neighbors. The segmentation process is divided into 3 main step : The initialization done with the K-mean algorithm The parameter estimation step, made with the iterative conditional estimation The segmentation step using the criteria of maximum a posteriori (MAP). The segmentation result is given on the NGC 4254 Cube, where each pixel belong to a class.
  • In the statistical image segmentation framework, we have to estimate the hidden states X from the measurement Y, with no determinist link between X and Y. We use the probabilistic approach where the probability of X and Y is defined by prior model, which can be global or local and an observation model. The estimation of the parameter can be computed by 3 algorithms EM, SEM or ICE, based on the maximization of the log-likelihood. The final segmentation step is done using either maximum a posteriori or the maximum a posteriori marginal criteria.
  • In our case, we have also labels and measurement field. The measurement are the multi-tracer scans acquired at different time. On the one hand the multi-tracer fusion, the estimation are computed with the measurement of different tracer acquired in the same time. On the other hand, in a patient follow-up context, the estimation are computed with measurement of 1 tracer acquired at different same time. In both cases, we have to estimate the X,Y distribution, with a mixture defined by the prior of X, and the mean value and the covariance matrix associate to each class.
  • Our fusion process is divided into three steps : The initialization with the K-means and the fuzzy-K means algorithm. The estimation of the X,Y distribution is realized by maximizing the log-likelihood. Thus, we have implemented the EM and the SEM algorithm, blind and adaptive version. The segmented map is obtain with the MAP criteria at the decision step. We have tested our algorithm on 2 kinds of data set, synthetic images and simulated tumors.
  • With synthetic images, we created maps of with 2 or 3 labels per images. The measurement, derived from the map of label was chosen mean or variance discriminate. First of all, we have fused images with the same map of labels. Thus the number of labels in the segmentation map will be equals to the number of labels in both images. And here are the segmentation results for 2 spectral bands and 2 labels per image. and 2 spectral bands and 3 labels per image.
  • Secondly, we have fused images with the different map of labels. Thus, there will have additional labels in the segmented map. The segmentation results for 2 spectral bands and 2 and 3 labels per image. In the segmented map, there is one new label. Then, we fused 2 spectral bands with 3 labels per image. The segmented map is composed of 5 labels. As you can see, the segmentation process failed with the AEM segmentation, contrary to the ASEM segmentation.
  • After the tests on synthetic images, we have tested the our method on simulated tumor. In this case, the fusion of tumor scan will be different The segmentation process should identify the new labels of the image. Each band is corresponding to a tracer scan, and each label to an level of uptake of the tracer. 3 cases are presented here : Fusion of 2 or three scans, with 2 or 3 levels of tracer uptake. In all cases, the initialization has been done with the fuzzy-K mean algorithm.
  • In the first and second cases, both AEM and ASEM well identify the different classes in the segmented map. In the third case, there is 7 effective labels. Both AEM and ASEM, correctly identify 6 classes, but failed in the segmentation of 1 classe.
  • Regarding our preliminary results, the fusion of synthetic images has been tested in different situations, give us satisfactory classification. Concerning the fusion of simulated tumor, the segmentation error depended on the number of classes in the label map, and the fuzzy K-means initialization. However, our segmentation process is not totally unsupervised, because the number of classes has to be defined by the user. For this preliminary results, the fusion process has been test on few spectral band.

Simon David Simon David Presentation Transcript

  • Multi-tracer analysis for patient’s following using multi-observation statistical image fusion : a feasibility study S. David 1 , M. Hatt 1 , P. Fernandez 2 , M. Allard 2 , O. Barrett 2 , D. Visvikis 1 1. LaTIM, INSERM U650, Brest, France 2. Department of nuclear medicine Hospital Pellegrin – CHU Bordeaux
  • Outline
    • Context of oncology
    • Positron Emission Tomography (PET)
    • External radiotherapy
    • Use of PET in clinical application
    • Multi-tracer analysis for dose-painting
    • Patient monitoring in PET
    • Multi-observation fusion
    • Multiband segmentation of a spectroscopic line data cube
    • Developed method
    • Preliminary results
    • Further work
  • Context of oncology
    • Cancer
    • In 2002 : 11 millions new cases and 7 millions deaths
    • Foresee in 2030 : 11 millions deaths
    • Diagnosis
    • Computed tomography (CT)
    • Magnetic resonance imaging (MRI)
    • Emission imaging (PET, SPECT)
    • Treatment
    • Surgery
    • Chemotherapy
    • Radiotherapy
    • Focus on the PET imaging and its application :
    • Radiotherapy planning
    • Therapy response assessment
    Since 2000 : PET / CT combined Gold standard for the diagnosis Usually combined
  • Context and motivations
    • Positron Emission Tomography (PET) :
    • Functional imaging : visualization of physiological processes
    • Mainly u sed for cancer diagnosis and staging
    • Principle :
    • Injection of a radiotracer
      • Biological tracer targets the tumor
      • Radionuclide : β - emitter
    • Detection of the 2 γ rays
    • Image reconstruction by tomography
    • Drawbacks of the PET imaging :
    • Blur (spatial resolution)
    • High noise (acquisition variability)
    • Low resolution ( >5 mm)
  • Context and motivations
    • PET/CT imaging :
    • Since 2000, acquisition of PET and CT in the same bed
    • Gold standard for the diagnosis
    + Combination of anatomical and functional information + Allow to anatomically locate the tracer uptake in the PET image - Registration of the CT and PET scans - Difference in the image resolution ( CT <1mm, PET >5 mm)
    • PET / CT used in the radiotherapy planning
    PET image CT image PET / CT fusion
    • Principle of radiotherapy :
    • Use the ionizing radiation to kill the malignant cells
    Context and motivations
    • External radiotherapy (most widely used) :
    • Photon or electron beams produced by a linear accelerator
    • Shape of the beam defined by the collimator
    • Definition of the target volumes :
    • Gross tumor volume (GTV)
      • Defined with conventional imaging modalities
    • Clinical target volume (CTV) > GTV
      • Volume computed considering the anatomical information
    • Planning target volume (PTV) > CTV
      • Take into account the physiological variations
    • Biological image-guided dose escalation
    • Development of Intensity-modulated radiation therapy (IMRT)
    • Administration of a non-uniform dose
      • Adapt the treatment to the patient
      • Reduce the irradiation of organ at risk (OAR) and surrounding healthy tissues
    • 18 F-FDG : Measure the glucose consumption (tumor highly glucose consumer)
      • Radiotracer the most widely use in PET
      • Not tumor specific (other physiological processes need glucoses)
      • Uptake in inflammatory tissues
    Use of PET in clinical application
    • Improvement of the tumor volume definition with the multi-tracer analysis
    Multi-tracer analysis IMRT planning PET / CT : Target volume definition
    • Development of specific radio-marker measuring ≠ tumor features :
    • FMISO : measure of the tumor hypoxia (lack of oxygen)
      • Hypoxia induces resilience to the radiotherapy
    Use of PET in clinical application
    • 18 F-FLT : measure of the tumor proliferation
      • Tumor specific radio-marker : ≠ FDG
      • Avoid inflammatory tissues
      • Lower uptake in the tumor than FDG
    Y. Yamamoto et al, European Journal of Nuclear Medicine and Molecular Imaging , 2008
    • Hypoxic tumors require a dose boosting
    • Merging all features measured by the tracers
    • Each radiotracer is measuring a specific biologic process
      • Scans are quantitatively not comparable
    FDG coronal PET scan FLT coronal PET scan
  • Use of PET in clinical application
    • Patient follow-up with PET :
    • Patients underwent chemo / radio-therapy
      • Early assessment of the response to the treatment
    • Adapt the therapy to the response
    • Avoid toxic and costly cares for non-responding patients
    • PET acquisition during the course of the therapy
      • Estimation of prognosis
    PET 1 pre-treatment PET 2 post-treatment Tumors
  • Goal of our work
    • Multi-sensor observation of an objet
    • Many clinical data available with the multi-modality imaging
    • Goal :
    • Fusing all the scans obtained :
      • with the ≠ radiotracers
      • Should improve the tumor volume definition
      • at different time of the treatment
      • A more accurate assessment of therapy response
    • Approach :
    • With the statistical segmentation framework
    • Model the information of multi-tracer and/or follow-up scans
    Analogy with the astronomical framework
  • Multiband segmentation of a spectroscopic line data cube
    • EM algorithm on Gaussian mixture model
      • fit the spectrum : assessment of mean and variance
    F. Flitti et al, Multiband segmentation of a spectroscopic line data cube : application to the H1 data cube of the spiral galaxy NGC 4254
    • Segmentation process based on Bayesian inference
    • Observation with radio interferometers
      • 3D data cubes (astronomical coordinates and frequency third axis)
    • Choice of the 6 most pertinent Gaussian
      • K-means
    • Computation of weight associated to each class
      • Levenberg-Marquardt algorithm
    Reduction of the dimension (42 channels) of the cube Maps of the weights of the 6 Gaussian with the NGC 4254 Cube
    • Bayesian classifier
    • Hierarchical Markovian Model
      • Models the spatial dependencies between neighbors
    • Results on the NGC 4254 Cube
    • Creation of a label map
    Multiband segmentation of a spectroscopic line data cube
    • Steps of the segmentation process
    • Initialization : K-mean algorithm
    • Parameter estimation step
      • Unsupervised with the Iterative Conditional Estimation (ICE)
    • Segmentation step
      • Criteria of maximum a posteriori (MAP)
    F. Flitti et al, Multiband segmentation of a spectroscopic line data cube : application to the H1 data cube of the spiral galaxy NGC 4254
  • Local prior model (spatial or contextual) Blind, contextual, adaptive… Observation model (noise) Gaussian, generalized gaussian Iterative estimation of the parameters deterministic (EM) stochastic (SEM) hybrid (ICE ) Multi-observation method
    • Probabilistic approach (Bayesian inference)
    Global Markovian model (field, quadtree…) Segmentation MAP, MPM criteria
    • Statistical image segmentation :
    • Estimation of (hidden) with (measurements)
    • No determinist link between and
    • X : labels field
    • Y : measurements field
    Multi-observation method
    • Estimation of the (X,Y) distribution
    • Mixture defined by θ =( α , β )
      • α i : priors of X
      • β i =(m i ,Г i ) : distribution of Y conditional to X
    Multi-tracer analysis Patient follow-up PET Tracer N Label X … … Label X Measures Y time PET MISO ... PET FDG … … … … … … … ... ... … … … …
  • Multi-observation method : preliminary work
    • Estimation of the (X,Y) distribution
    • Maximization of the log-likelihood
    • Implementation of EM and SEM algorithm
      • Blind and adaptive (AEM,ASEM) version
    • Initialization :
    • K-means
    • Fuzzy K-means
    • Test on data set :
    • Synthetic images
    • Simulated tumors
    • Decision step
    • MAP criteria
          • Creation of a segmented map
    Fusion process :
  • Preliminary results
    • Synthetic images :
    • Map of labels X
      • 2 or 3 labels per image
    • Measurements Y
      • Mean discrimination (MD)
      • Variance discrimination (VD)
    • Fusion of measurement with the same map of labels
      • Number of labels in segmentation map = Number of labels per image
    Map of labels X
    • N = 2 spectral bands
    • K = 3 labels
    • Random initialization
    AEM segmentation ASEM segmentation VD MD
    • Segmentation results :
    • N = 2 spectral bands
    • K = 2 labels
    • Random initialization
    VD AEM segmentation ASEM segmentation Measurement Y MD Segmentation map
  • Preliminary results
    • Synthetic images :
    • 2 or 3 labels per image
    • Fusion of measurement with the different map of labels
      • Additional labels in the segmented map
    AEM segmentation ASEM segmentation
    • N = 2 spectral bands
    • K = 3 labels per scan
    • 5 classes in segmentation map
    • Fuzzy K-means initialization
    AEM segmentation ASEM segmentation Measurement Y Segmentation map
    • Segmentation results :
    • N = 2 spectral bands
    • K = 2 and 3 labels
    • 4 classes in segmentation map
    • Fuzzy K-means initialization
  • Preliminary results Simulated tumors :
    • The ground truth of each tumor scan will be different
    • The segmentation process should identify the new labels
    • Each band : a tracer scan
    • Label in the scan : uptake of the tracer
    3 cases :
    • N = 2 tracer scans
    • K = 2 levels of tracer uptake per scan
    • Fuzzy-K mean initialization
    • N = 3 tracer scans
    • K = 2 levels of tracer uptake per scan
    • Fuzzy-K mean initialization
    • N = 2 tracer scans
    • K = 3 levels of tracer uptake per scan
    • Fuzzy-K mean initialization
  • Preliminary results Segmentation results :
    • N images = 2 , K tracer uptake =2
    • N images = 3 , K tracer uptake =2
    • N images = 2 , K tracer uptake =3
    AEM segmentation ASEM segmentation AEM segmentation ASEM segmentation AEM segmentation ASEM segmentation
    • Fusion of the synthetic images
    • In the different situations (N Bands, K classes)
      • Supervised, semi-supervised and unsupervised segmentation
      • Satisfactory classification
    • Fusion of simulated tumors
    • Segmentation error depends on :
      • The fuzzy K-means initialization
      • Noise level in the scans
      • The number of classes in the label map
    • Limitations :
    • Segmentation not totally unsupervised :
      • Number of labels has to be defined by the user
    • Fusion of few spectral bands
    Preliminary results
  • Further work
    • Fusion process with more data
      • Other tracers images and / or follow-up scans
    • Segmentation totally unsupervised
      • Estimation of the classes number in the label map
    • Test the method on simulated data with GATE
      • More realistic simulated tumors
      • Computation of classification error
    • Application of our method in a radiotherapy planning station
  • Thank you for your attention