This document describes a study using wavelet transform to enhance contrast in mammographic images of a phantom. The methodology involved developing a mammographic phantom with anatomical features, acquiring images using a digital mammography system, preprocessing the images using Gaussian filtering, enhancing the images using biorthogonal 2.8 wavelet filtering, and having observers interpret and score the original and enhanced images in a receiver operating characteristic analysis. The results showed that wavelet transform enhancement improved detection of micronodules but did not significantly improve detection of nodules or fibrils compared to the original images. Wavelet transform was found to reduce noise and improve contrast.

1. MAMMOGRAPHIC PHANTOM IMAGES
CONTRAST ENHANCEMENT
USING
BIORTHOGONAL 2.8 WAVELET FILTER
By:
Nor’Aida Binti Khairuddin
Physics Department, Faculty of Science, UTM

2. PROBLEM STATEMENTS
 Most mammographic image have very
small features, poor contrast, noisy
and wide range of anatomical
patterns.
 Missed or misinterpreted by the
radiologists.
 Increase false positive cases.

3. Development of the
mammographic phantom with
anatomical features.
Receiver Operating
Characteristic (ROC ) Analysis
Image acquisition
Image dataset
Image pre
processing
Image
enhancement
Image scoring
METHODOLOGY

4. METHODOLOGY
Development of the mammographic
phantom.
The mammographic phantom
template contains fibrils (nylon string),
micronodules (SiO2) and nodules
(wax) arranged randomly as an
alternative method for evaluating
imaging systems.

5. METHODOLOGY
Image Acquisition
 Mammographic phantom was
located at the bottom of the
perspex (acrylic) with the same
size.
 Mammographic images were
obtained using Hologic Lorad
Selenia Full Field Digital
Mammography System using AEC
function.
 kVp range 28 kV to 30 kV and Rh
(Rhodium) as a filter.

6. METHODOLOGY
Image Dataset
 The original raw mammographic
phantom image obtained of 29 kVp
and 124.6 mAs and 5.8 cm
compressions stored in DICOM
format.
 Before pre-processing the images,
the DICOM mammographic
phantom images were converted to
TIFF (Tagged Image File Format)
format.

7. METHODOLOGY
Image pre-processing
 Denoised image using low pass Gaussian filter size 30 × 30 and
standard deviation,  = 0.6

8. METHODOLOGY
Image Enhancement
 Biorthogonal 2.8 wavelet filter
filter by decomposing the sub
band transformation in 2D
wavelet transform.
Figure: Displaying the wavelet decomposition
coefficients structures by two level
decomposition of Biorthogonal 2.8 wavelet
filter.

9. METHODOLOGY
Image Scoring
 Observer interpreted mammographic
images subjectively.
 Based on 5 confidence levels.
 1 : definitely not present
 2: probably not present
 3: no decision possible
 4: probably present
 5: definitely present

10. METHODOLOGY : ROC ANALYSIS
 The operating points based on subjective interpretation score were calculated
using Microsoft Excel.
 CORROC2 software used to process the clustered data from the ROC scoring and
operating point calculation dataset.

11. METHODOLOGY: ROC ANALYSIS
 ROC curve fitting and
statistical testing from
data collection to compare
two detection modalities.
 Sensitivity (True Positive
Fraction)
 False Positive Fraction
(FPF) = 1 – specificity
 Area index of curves (Az)
were compared between
A and B by determining
the  values.
  values used to
determine the significance
of the differences of area.
FPF

12. RESULTS AND DISCUSSION
Detection of Nodules
Detection of Nodules
Figure : ROC curves for detection of nodules from original and wavelet transform
enhanced images.

13. RESULTS AND DISCUSSION
Detection of Fibrils
zDDetection of Fibrils
Figure : ROC curves for detection of fibrils from original and wavelet transform
enhanced images.

14. RESULTS AND DISCUSSION
Detection of Micronodules
Detection of Micronodules
Figure : ROC curves for detection of micronodules from original and wavelet
transform
enhanced images.

15. DISCUSSIONS
 The  values for detection of nodules, fibrils and micronodules were
larger than 0.05. The statistical data for all detection were not statically
significance.
 Detection of nodules have the lowest area index values (Az < 0.97).
 Wavelet transform enhancement improved the detection of
micronodules with higher sensitivity and lowest false positive rates.
Micronodules have higher mass attenuation coefficient value.
 Wavelet transform enhancement could reduce noisy pixels of image and
improved image contrast.

16. Thanks to Malaysian Ministry of Science, Technology and Innovation (MOSTI) and
Universiti Teknologi Malaysia (UTM) for their financial funding through Science
Fund grant.