Svm based cbir of breast masses on mammograms

SVM-based CBIR of Breast

Masses on MammogramsL. TSOCHATZIDISK. ZAGORISM. SAVELONASI. PRATIKAKIS

Democritus University of Thrace

Department of Electrical and Computer Engineering

Visual Computing Group

August 17, 2014

European Conference on Artificial Intelligence (ECAI) 2014 ,

Workshop on Artificial Intelligence and Assistive Medicine (AI-AM)

1

Mammography

Diagnostic and screening tool of

breasts

Dominant imaging modality for early

detection of breast cancer

Breast cancer appears as a mass

and/or microcalcifications

The diagnosis is difficult that leads to

unnecessary biopsies

August 17, 2014European Conference on Artificial Intelligence (ECAI) 2014 ,

Workshop on Artificial Intelligence and Assistive Medicine

(AI-AM)

2

Computer Aided Decision (CAD)

Consists of two sub-categories:

Systems for detecting an abnormality - Computer Aided Detection (CADe)

Systems for diagnosing an abnormality - Computer Aided Diagnosis (CADx)

CAD systems usually employ classification schemes for benign-malignant

discrimination



(AI-AM)

3

CBIR-CAD Systems

CAD systems that incorporate a CBIR step prior to decision

Retrieve similar images based on low-level image features

Provide visual aid, enables consulting previous cases, leading to

increased confidence into incorporating CAD-cued results



(AI-AM)

4

Problem Definition – Classes (1)

Circumscribed Smooth and highly convex boundary

Well-defined margin

Low probability of malignancy



(AI-AM)

5


Microlobulated Rough and bumpy boundary

The overall shape is retained

Medium to high probability of

malignancy



(AI-AM)

6


Spiculated Margin with large protrusions and not

clearly defined

The overall shape becomes irregular

Highly suggestive of malignancy



(AI-AM)

7

CBIR-CAD’s pipeline



(AI-AM)

Lesion Segmentation

Mass Detection

CBIR DiagnosisInput Image

BENIGN

8

CBIR Architecture



(AI-AM)

9

Feature Extraction – Global Shape (1)

Solidity factor: The degree that the shape deviates from its convex hull

𝑆𝑜𝑙𝑖𝑑𝑖𝑡𝑦 =𝐴𝑟𝑒𝑎 𝑜𝑓 𝑚𝑎𝑠𝑠 (𝐴)

𝐴𝑟𝑒𝑎 𝑜𝑓 𝑖𝑡𝑠 𝑐𝑜𝑛𝑣𝑒𝑥 ℎ𝑢𝑙𝑙 (𝐻)

Compactness factor: The degree that a shape deviates from a perfect

circle

𝐶𝑜𝑚𝑝𝑎𝑐𝑡𝑛𝑒𝑠𝑠 = 1 −4𝜋𝐴2

𝑃2



(AI-AM)

10

Feature Calculation – Global Shape

(2)



(AI-AM)

Circumscribed Microlobulated Spiculated

11

Feature Extraction – DFT of NRL

Normalized Radial Length Function

1. The distance of each contour point

to the shape’s center of gravity

2. Normalized by the average radial

length

3. Computation of Discrete Fourier

Transform coefficients



(AI-AM)

12

CBIR Architecture



(AI-AM)

13

The SVM Layer – Support Vector

Machines (1)

Binary Linear Classifiers

For non-linear problems: Projection of

samples to a higher dimensionality

space.

Finds a hyper-plane that optimally

separates the two classes

Decision function:

𝑓 𝑥 = 𝑠𝑖𝑔𝑛(𝑤 ∙ 𝑥 + 𝑏)



(AI-AM)

14

The SVM Layer – Structure

An ensemble of binary SVM classifiers

is employed

One SVM for each class – Three SVMs

in total

Each SVM outputs the participation

level of a sample in the corresponding class



(AI-AM)

15

The SVM Layer – Participation value

Computation

SVM Decision Function

𝑓 𝑥 = 𝒘 ∙ 𝒙 + 𝑏

Proposed normalization

𝑅 𝑥 =

𝑚𝑎𝑥1

1 +13𝑒𝑓(𝑥)

,1

1 +13𝑒−𝑓(𝑥)

, 𝑖𝑓 𝑓 𝑥 > 0

1 − 𝑚𝑎𝑥1

1 +13 𝑒

𝑓 𝑥,

1

1 +13 𝑒

−𝑓 𝑥, 𝑖𝑓 𝑓 𝑥 < 0



(AI-AM)

16

Experimental Results

Experiments on a dataset of total 90 mammograms (CC views) from DDSM

Manual contour delination from expert radiologist

Equal number of mammograms from each class

The 2/3 of dataset was used for the SVM training

The Rest 1/3 was used as test set

Comparison between proposed method and the typical, unsupervised one.



(AI-AM)

17

Experimental Results – Evaluation

metrics

Precision at N (P@N): The percentage of correct images at the top-N

places of the rank list (N=5)

Mean Average Precision (MAP): Measures the overall performance of a

query

𝐴𝑃 = 𝑘=1𝑛 𝑃@𝑘 ∗ 𝑟𝑒𝑙 𝑘

𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑟𝑒𝑙𝑒𝑣𝑎𝑛𝑡 𝑑𝑜𝑐𝑢𝑚𝑒𝑛𝑡𝑠𝑟𝑒𝑙 𝑘 =

1, 𝑖𝑓 𝑘_th 𝑖𝑚𝑎𝑔𝑒 𝑖𝑠 𝑐𝑜𝑟𝑟𝑒𝑐𝑡0, 𝑒𝑙𝑠𝑒

𝑀𝐴𝑃 =1

𝑁

𝑁

𝐴𝑃



(AI-AM)

18

Experimental Results

Classes Unsupervised CBIR Supervised CBIR

P@5 MAP P@5 MAP

Circumscribed 0.90 0.91 0.90 0.92

Microlobulated 0.71 0.72 0.71 0.76

Spiculated 0.65 0.61 0.74 0.73

Average 0.75 0.74 0.78 0.80



(AI-AM)

19

Experimental Results – Circumscribed



(AI-AM)

20

Experimental Results – Microlobulated



(AI-AM)

21

Experimental Results – Spiculated



(AI-AM)

22

Conclusions

CBIR system for retrieval of masses on mammograms

Evaluation showed that the features have high discriminant ability

The supervised CBIR offers enhanced results as compared to the

unsupervised one.

The final vectors used are very small compared to the initial feature

vectors



(AI-AM)

23

Future Work

Test the CBIR scheme for microcalcifications

Integration of CBIR within the context of a complete mammographic CAD

system



(AI-AM)

24

Thank you!

Any questions?



(AI-AM)

25

Svm based cbir of breast masses on mammograms

Science

artificial intelligence

european conference

assistive medicine aiam3cbir

assistive medicine aiam7cbir

cad systems cad systems

overall shape

shape deviates

cadcued resultsaugust