Transcript Breast Cancer Diagnosis

Breast Cancer Diagnosis
A discussion of methods
Meena Vairavan
Presentation Outline




Problem Statement
Description of Data
Methods of Diagnosis
Future Plans
Problem Statement
My goal is to compare two computational
methods to determine which is a more
effective means for performing breast
cancer diagnosis. The first method
uses linear programming and the
second method uses neural networks.
Both methods analyze data generated by
fine needle aspiration tests.
Description of Data

Source of Data: Wisconsin Diagnosis Breast Cancer
Database (WBCD)
– Dr. William Wolberg - Department of Surgery
– Professor W. Nick Street - Department of Manag. Sciences
– Professor O.L. Mangasarian - CS Department


Each case is represented by a 30-dim. feature vector
computed from a digitized fine needle aspirate of a
breast mass.
The features describe characteristics of the cell
nuclei present in the image.
– Radius, texture, smoothness, concavity, and symmetry
Methods of Diagnosis


Method 1: Diagnosis through linear
programming via generation of a
separation plane.
Method 2: Diagnosis through the use
of a multi-layer perceptron model using
back propagation techniques.
Diagnosis Through LP

A linear function was constructed to
generate a separation plane to classify
malignant and benign tumors.
– f(x) = ’x - 
– f(x) > 0 for malignant cases
– f(x) < 0 for benign cases
– minimize misclassified points by choosing
 and  to minimize distance from f(x)
Future Plans

Find the optimal MLP configuration for
this diagnosis.
– Plan to modify Professor Hu’s back
propagation method for these purposes.


Choose appropriate characteristics to
compare the LP method and MLP
method
Provide an analysis of my results.