Home


Breast Cancer Prognosis Using Image Processing

Main


Overview Abstract Main Results Future References Download Terminology Links About Contact
As mentioned before the concept of this study is the validity of using a computer in the prognosis of breast cancer.

This project is divided into 3 stages:

  1. The location of important features i.e. Cells and the Cells Nucleus
  2. Calculate Prognostic Values
  3. Classification & Testing
Stage 1 The location of important features
This involves segmenting the cells from the background, below is an image containing breast cells.
image13.jpg
As can be seen, it could be quite difficult for a person to determine by eye where the boundaries of some cells are. The second stage entails quantifying prognostic indicators such as nuclear size, so to enable an accurate calculation this stage (segmentation) needs to be as accurate as possible. If the segmentation is of poor quality then subsequent calculations will be of no use.

The method of segmentation that I chose to develop is based on the histogram of the image but there are alternatives such as (a) the Hough transform and (b) ************.
The link below will take you to a page depicting the image processing algorithm that automatically segments the Cells Nucleus from the background.

Click here for Segmentation algorithm which was implemented using an Image Processing Toolkit from WIT

Select below for example images processed by the Segmentation Algorithm.

Example 1
Example 2

The algorithm is relatively fast, low memory intensive and automatically detects a high percentage of Nuclei within a given sample.
Data collected about the detected Nuclei was then written to a file so as to quantify prognostic indicators, as shown below.

Filename Description
xCentroids.txt the x location centers of each detected object
yCentroids.txt the y location centers of each detected object
nPixels.txt The number of pixels contained within the perimeter of each detected object.
perimLength.txt The length in pixels of the perimeter of each detected object
points.txt The perimeter locations in (x,y) format of each detected object.

These data files contain the required information for us to quantify the prognostic indicators.

Stage 2: Quantification of prognostic indicators
The prognostic indicators used in this study are as follows (for a more details regarding prognostic indicators refer to section 2 of the project report - Literature Survey which is accessible via the download page).
  1. Nuclear size
    The number of pixels of each detected object are summed, the mean size is then calculated.
  2. Cell uniformity
    This measured by taking the Standard Deviation of the compactness ratio.
  3. Cell dissociation
    This indicator was quantified by calculating the mean of the distances between the center of each cell to its nearest neighbour.
    Distance = sqrt ((x2 - x1)sqr + ( y2 - y1 )sqr )
  4. Nuclear shape
    This indicator can be determined by calculating the mean compactness ratio, as follows:
    compactness = area/(perimeter length)sqr
The segmentation algorithm was applied to each image in the data set of which there were 30. The prognostic indicators were then calculated for each image. An example of the the quantified prognostic indicators can be seen below.

Image Nuclear Size Nuclear Shape Uniformity Dissociation
4 228.414 19.0340 11.3013 36.8793
12 200.718 18.9457 7.50105 37.8169


Stage 3: Classification & Results

The cells that were the study of this project have already been graded using the Nottingham Prognostic Index, this is a well known and respected prognostic indicator. So now there exists an exact value that represents a prognosis and 4 prognostic values, all tha remains is to test to see if a correlation exists between the calculated prognostic values and the known prognosis.

Some of the data that has been collected for each image will be used to build a classifier, this classifier will then be tested on the remaining set of data, this is known as cross validation. The system used to build the classifier is called C5
The data file containing the calculated prognostic values along with corresponding NPI scores can be viewed here .

NEXT

  Home | Overview | Abstract | Main | Results | Future | References | Download  | Terminology  | Links  | About The Author | Contact us 

Copyright © 2002 Nick Bryan All rights reserved.



Hosted by www.Geocities.ws

1