Mammogram%20Analysis%20

1
Mammogram Analysis Tumor classification

• - Geethapriya Raghavan

2
Background

• Mammogram
• X-Ray image (of gray levels) of inner breast
  tissue to detect cancer
• Shows the levels of contrast characterizing
  normal tissue and vessels
• Issues
• Detect abnormalities (tumors)
• Diagnosis - Classify as benign or malignant
• Remove noise

3
Microcalcifications
Mammograms obtained from MIAS database
4
Methods ..

• Non-linear classifiers preferred over linear
  classifiers given the randomness in occurrence of
  tumor cells
• Contemporary methods - supervised learning
  problem (Wei et al., 2005)
• Support Vector Machines (SVM) (Vapnik et al.,
  1997)
• Kernel Fisher Discriminant (KFD)
• Relevance Vector Machines (RVM)

5
Method I - SVM

• SVM was used by Chang et al., on US images
• Texture feature microcalcification area,
  contrast.
• Software SVM Light ((http//svmlight.joachims.or
  g/)
• The best fitting hyperplane f(x) wT . x b
  forms the boundary
• For non-linear SVM, the x in the above equation
  is replaced by a nonlinear function of x.

6
Method II

• Use of wavelet transform to decorrelate data
  (image) (Borges et al., 2001)
• Obtain wavelet coefficients as features
• Normalize coefficients and feed into Nearest
  Neighborhood classifier
• Wavelet decomposition - Low frequency
  coefficients extracted at two levels and NNR run
  with euclidean distance as metric.

7
Results
Classifier Microcalcification Contrast Microcalcification Area
Non-linear SVM 67.7 78
Linear SVM 42.8 70.4
NNR 72 76.2
8
Results - ROC

• Sensitivity Number of True Positive
  Classifications
• Number of
  Malignant Lesions
• Specificity Number of True Negative
  Classifications
• Number
  of Benign Lesions
• Sensitivity (y) vs. Specificity (x)
• Dotted lower bound
• Red line Wavelets NNR
• Black curve linear SVM