The CALMA project

1
The CALMA project

• A CAD tool in breast radiography
• A.Ceccopieri, Padova 9-2-2000

2
Computer Assisted Library in MAmmography
Screening mammography sensitivity (identified
positives / true positives) 73 -
88 specificity (identified negatives / true
negatives) 83 - 92 These merit figures
INCREASE if diagnosis is performed by 2
independent radiologists
3

• CALMA aims to
• Build a DATABASE of mammograms in digital format
• Perform an automatic classification of parenchyma
  structures
• Detect the spiculated lesions
• Detect micro-calcification clusters

4
FA 37
OUR DATABASE
DN 5
900 patients 2900 images
Glandular 58
5
HARDWARE
DAQ granularity 85 mm range12 bit dimensions
2000x2600 pixels
STORAGE 60 images/ CD (no compression) up to 240
CD
6
DAQ panel database search
Preview and images description
Queries
Full screen display
7
Automatic classification of breast parenchyma
Spatial frequencies analysis (FFT)
Left to right / top to bottom - dense (DN) -
irregularly nodular (IN) - micro-nodular (MN) -
fiber-adipose (FA) - fiber-glandular (FG) -
parvi-nodular (PN) -Glandular (INMNFGPN)
Supervised FF-ANN
8
2dim FFT
Feature extraction
512x512 pixels analysis
ANN classification
GLANDULAR
9
RESULTS TEXTURE ANALYSIS
DENSE
ADIPOSE GLANDULAR DENSE
gt95 0
0 ADIPOSE 16
683 16
GLANDULAR 4 3
931
10
SPICULATED LESIONS
Unroll spirals Spatial frequencies
analysis(FFT) FF-ANN
examples
11
RESULTS _at_ sensitivity90(3)
Method Area (cm2) spread (cm2) B (0-0) 31 16 B
(1-3) 27 13 B (2-5) 25 13 C neural 36 12 C
normalized 36 18 C corona 49 27
12
Integration range 2-5
13
Spiculated lesions CAD performances
Red radiologist
Blue CAD
14
RESULTS SPICULATED LESIONS
Sensitivity (per patient) 903 FALSE POSITIVES
/ IMAGE 1.4 AVERAGE ROI 25 cm2 DATA REDUCTION
10
15
MICROCALCIFICATION CLUSTERS
FF-ANN Sanger learning rule
PCA
Examples
16
Method

• Image Preprocessing (convolution filters)
• PCA through a NN trained with the Sanger rule
• Study of the first Principal Components
• Classification

17
Preprocessing

• 60x60 pixels windows selection
• convolution filters with dims 5x5 7x7 9x9

Best results with a 7x7 filter with A1\N2 ??
aij lt0 (aij kernel element)
18
Results
No Micro-calcification clusters
With micro-calcification clusters
Sensitivity 73 2 Specificity 94
2
19
Micro-calcification clusters CAD
Red radiologist
Blue CAD
3
2
1
20
RESULTS MICRO-CALCIFICATION CLUSTERS
SENSITIVITY 732 SPECIFICITY 942
21
FUTURE

• Software developement 1- Local
  classification of parenchyma
  2- Use parenchyma classification
  for lesions CAD
  3-
  Use the asymmetry between the two sides to detect
  cancer.
• Increase the DATABASE
• ON-LINE Validation Is CALMA a good (second)
  radiologist?
• Implementation of physician-friendly CAD
  workstations in the collaborating Hospitals