Brain Tumour Analysis Project

Document Sample
Brain Tumour Analysis Project Powered By Docstoc
					Brain Tumour
Analysis Project
                   Introduction

 Brain Tumours
   A brain tumour is a mass of abnormal cells within or
    around the structure of the brain and can be benign or
    malignant, or be primary or secondary.
   Malignant Tumours are of the greatest concern
    because they can be fast growing, have poorly
    defined borders, and can invade surrounding tissue
    and structures.
   A primary brain tumour originates from cells within the
    brain, while a secondary brain tumour originates
    somewhere else in the body and then travels to the
    brain.
Magnetic Resonance Imaging
(MRI)
      > Magnetic resonance
        imaging (MRI) is a
        method to produce
        high quality images of
        the head and brain.
      > Radiologists can
        locate cancer on the
        MRI and attempt to
        remove it.
> Radiation based treatment consists of
  applying conformal radiation therapy,
  which is the use of radiation by
  delivering an appropriate dose to the
  cancer tissue by irradiating the patient
  from different directions with high energy
  photon beams.
                 What is an MRI?

> An MRI can be thought of as taking an image
  slice through the human body and each one of
  the slices has a thickness.
> This form of imaging is in some respects
  equivalent to cutting off the anatomy above the
  slice and below the slice. The slice is said to
  be composed of several volume elements or
  voxels.
> The volume of a voxel is approximately 3 mm3.
                 What is an MRI?

> The following figure shows three types of
  images that can be taken of each slice.
> T1 highlights fat locations.
> T1c uses a contrast agent (gadolinium) to
  highlight tumours.
> T2 highlights water locations.
                                        What is an MRI?




T1 weighted: An MRI that highlights   T1c weighted: An MRI that is taken          T2 weighted: An MRI that highlights
fat locations.                        after the injection of the contrast agent   water locations.
                                      gadolinium, a contrast agent that can
                                      make abnormalities such as tumors
   T1 Tissue           How it         clearer due to the element's special                              How it
                      appears         magnetic properties.                          T2 Tissue
                                                                                                       appears
      Bone              Dark                                                           Bone              Dark
       Air              Dark                                                            Air              Dark
       Fat              Bright                                                          Fat              Dark
      Water             Dark                                                           Water             Bright
  Case Study 1

> The following case shows how
  the MRI is used to assist in
  tumour treatment.
         Case Study 1: Raw data T1, T1c, T2




 T1                 T1c            T2
Notice how the contrast agent
gadolinium reveals the tumour.
             Problem

> MRIs can only detect what is called the
  gross tumour volume (GTV) which is the
  abnormal tumour region visible in the MRI.
> Some cells, which grow from the tumour's
  surface as it spreads, are invisible on
  MRIs. These are referred to as “occult”
  cells (“meaning hidden”).
               Problem

> Historically radiation oncologists have had
  to assume the hidden cells are adjacent to
  the visible tumour.
> Because they don‟t know where they are,
  they have to treat every adjacent area as
  equally probable of harbouring hidden cells.
               Problem

> Currently, oncologists have computers grow
  the GTV approximately 2 cm larger in all
  three dimensions which gives them a
  spherical volume (or envelope) around the
  tumour that they can treat.
> Unfortunately, this means a lot of cells that
  are healthy are also irradiated.
         Case Study 1: GTV with 2 cm
         Envelope




Notice how much larger the
envelope is than the tumour.
                 Hypothesis

> Because the tumours are not round and have
  a shape and a direction to them, researchers
  believe that growing the tumour 2cm in every
  direction is killing more cells than necessary
  and that some hidden cells may grow outside
  the 2cm envelope.
> Their hypothesis is that there must be a
  preferred direction for the tumour growth and
  if there is a preferred direction, it can be
  determined through Machine Learning
  techniques.
                 Hypothesis

> Machine Learning is a subfield of artificial
  intelligence, and is concerned with the
  development of algorithms and techniques
  that allow computers to "learn".
> The goal of Machine Learning in this project is
  to produce an accurate volume of tumour
  automatically. That is, without any human
  assistance.
> This is done through computational and
  statistical methods.
                 Procedure

> Researchers are locating tumour volumes
  within the patient's brain.
> The process is called an „Automated
  Segmentation Program‟ (ASP) and its
  purpose is to let computers classify the best
  possible volume of area to be treated.
> For each patient there are 20 axial slices for
  the brain starting from the neck up to top of
  the head for each of three modalities: T1,
  T1c, T2.
                > This shows 5 slices taken of
                  the same patient.
                > Each row corresponds to an
                  axial slice of the patient at
                  different heights.
                > The first 3 columns are
                  weighted T1, T1c, and T2.




T1   T1c   T2
                    Case Study 1: Automated
                    Segmentation Program (ASP)




The computer has learned a        However, one area has
different classification than     also been chosen that
the 2 cm envelope.                is not a tumour.
Case Study 1: Automated
Segmentation Program (ASP)




        Incorrect classifications (called
        an outlier) are removed and
        edges are “smoothed”.
                   Case Study 1: ASP Envelope and 2
                   cm Envelope displayed




Results:                          Results:
• 2 cm envelope shown in red.     • Orange area is where
• ASP envelope shown in yellow.   envelopes are the same.
Red = 2 cm envelope                 Yellow = ASP envelope




                      Orange = Chosen by both techniques
                 Questions

1. What is the difference between a voxel in an
   MRI and a pixel from a computer image (for
   example, a jpeg, bitmap, etc)?
2. Why do oncologists want to minimize the
   amount of brain tissue treated by irradiation?
3. Why is growing a 2cm envelope all around
   the tumour, in every direction, not
   necessarily the optimal treatment?
4. Why is a malignant brain tumour more of a
   danger than a benign one?
                Questions

5. In the case study shown above the total area
   selected by the 2 cm envelope and by the
   ASP envelope appear to be almost equal.
   Why do researchers believe the ASP
   envelope is more accurate?
6. For Machine Learning techniques to be
   effective, the MRIs must be consistently
   imaged. What are some possible variables
   that may affect the results?
               Questions

7. How is Machine Learning different from
   programming that uses expert knowledge?
8. Look closely at the MRIs in Case Study 1.
   What do you notice about the symmetry of
   the brain? Why do you think this might be?
9. The goal of Machine Learning techniques is
   to have computers learn on their own. Once
   they can do that, what will be the role of
   people in cancer treatment?
For more information on the Brain Tumour
   Analysis Project, visit:

http://www.cs.ualberta.ca/~btap/index.php
Centre for Mathematics Science and Technology Education (CMASTE)
382 Education South
University of Alberta
Edmonton AB T6G 2G5
www.CMASTE.ca
To download: select Outreach, Alberta Ingenuity Resources and Centre for Machine Learning
Filename: AICML6BrainTumourAnalysis




Centre for Machine Learning
Department of Computing Science
University of Alberta
2-21 Athabasca Hall
Edmonton AB T6G 2E8
(780) 492-4828
www.machinelearningcentre.ca


Alberta Ingenuity
2410 Manulife Place, 10180-101 Street
Edmonton AB T5J 3S4
(780) 423-5735
www.albertaingenuity.ca