Title: A Computer Aided Detection System For Mammograms Based on Asymmetry and Feature Extraction Techniques
1A Computer Aided Detection System For Mammograms
Based on Asymmetry and Feature Extraction
Techniques
- By Mohammed Jirari
- Benidorm, Spain
- Sept 9th, 2005
2 Why This Project?
- Breast Cancer is the most common cancer and is
the second leading cause of cancer deaths - Mammographic screening reduces the mortality of
breast cancer - But, mammography has low positive predictive
value PPV (only 35 have malignancies) - Goal of Computer Aided Detection (CAD) is to
provide a second reading, hence reducing the
false positive rate
3Basic Components of the System
- Mammogram Normalization
- Mammogram Registration
- Mammogram Subtraction
- Feature Extraction
- Morphological Closing
- Morphological Opening
- Size Test
- Border Test
- ROC Analysis
4What is a Mammogram?
- A Mammogram is an x-ray image of the breast.
Mammography is the procedure used to generate a
mammogram - The equipment used to obtain a mammogram,
however, is very different from that used to
perform an x-ray of chest or bones
5Mammograms (cont.)
- In order to get a good image, the breast must be
flattened or compressed - In a standard examination, two images of each
breast are taken one from the top (CC) and one
from the side (MLO)
6Mammogram Examples
Mammogram of a left breast, cranio-caudal (from
the top) view
Mammogram of a left breast, medio-lateral oblique
(from the side) view
7Purpose of CAD
- Mammography is the most reliable method in early
detection of breast cancer - But, due to the high number of mammograms to be
read, the accuracy rate tends to decrease - Double reading of mammograms has been proven to
increase the accuracy, but at high cost - CAD can assist the medical staff to achieve high
efficiency and effectiveness - The physician/radiologist makes the call not CAD
8Proposed Method
- The proposed method will assist the physician by
providing a second opinion on reading the
mammogram, by pointing out area(s) that are
different between the right and left breasts - If the two readings are similar, no more work is
to be done - If they are different, the radiologist will take
a second look to make the final diagnosis
9Data Used
- The dataset used is the Mammographic Image
Analysis Society (MIAS) MINIMIAS database
containing Medio-Lateral Oblique (MLO) views for
each breast for 161 patients for a total of 322
images - Each image is
- 1024 pixels X 1024 pixels
10Normalization
The images were corrected/normalized to avoid
differences in brightness between the right and
left mammograms
11Mammogram Registration
- Thermodynamic concepts are used
- Match a model M with a scene S (M must be
deformed to resemble S as much as possible) - Use diffusion process technique as follows
12Mammogram Registration (cont.)
- 1. Select pixels to be demons
- 2. For each demon, store displacement then
apply Gaussian filter - 3. Use trilinear interpolation to estimate
intermediate intensities - 4. The demon force is given by optical flow
13Registration Example
Mammogram of left breast
Mammogram of right breast
14Registration Example (cont.)
Grid of displacement
Registered images
15Mammogram Subtraction
- Simple linear subtraction is used
- Flipped right left
- Most common gray level is 0
- Masses in right breast are in lower gray level
region of subtraction image histogram, while left
breast masses are in the higher gray level region
16Mammogram Subtraction Example
Flipped right breast
Left breast showing mass
17Mammogram Subtraction Example (cont.)
Subtraction image
Superimposed subtraction image
18Feature Extraction
- Many features are not masses
- Morphological filtering using a 3X3 kernel
- Size test (100 pixels)
- Border test for border misalignment
19Avg. of areas after each stage of the detection
process
Stage in detection process Avg. of detected areas
After subtraction 13.65
After morphological filtering 7.80
After size test 5.42
After border test 2.17
20Results
Recognition 93
False positive 1.26
TPF 0.9605
FPF 0.0962
Az 0.95
- 102 registered pairs of mammograms used
- Verified by expert radiologists
21ROC curve showing Az0.95
22Future work
- Use more features like brightness and
directionality - Try and reduce False Negatives on the basis of
region characteristics size, difference in
homogeneity and entropy - Use larger database that contains both MLO and CC
to train/learn, since most commercial CADs use
hundreds of thousands of mammograms to try and
recognize foreign samples
23 24Questions