the calma project
DESCRIPTION
The CALMA project. A CAD tool in breast radiography A.Ceccopieri, Padova 9-2-2000. C omputer A ssisted L ibrary in MA mmography. Screening mammography sensitivity (identified positives / true positives) 73% - 88% specificity (identified negatives / true negatives) 83% - 92% - PowerPoint PPT PresentationTRANSCRIPT
The CALMA project
A CAD tool in breast radiography
A.Ceccopieri, Padova 9-2-2000
CComputer omputer AAssisted ssisted LLibrary in ibrary in MAMAmmographymmography
Screening mammographysensitivity (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
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
900 patients
2900 imagesGlandular58 %
DN5 %
FA37 %OUR DATABASE
DAQ: granularity: 85 m
range:12 bitdimensions: ~2000x2600 pixels
STORAGE60 images/ CD (no compression)
up to 240 CD
HARDWARE
DAQ panel& database
search
QueriesFull screen display
Preview and images’ description
Automatic classification of breast parenchyma
Left to right / top to bottom:
- dense (DN)- irregularly nodular (IN)- micro-nodular (MN)- fiber-adipose (FA)- fiber-glandular (FG)- parvi-nodular (PN)
-Glandular (IN+MN+FG+PN)
0 2 4 6 8 10 12 14 16
0
20
40
60
80
100 Trasformata di Fourier
F(k)
k
SupervisedFF-ANN
Spatial frequencies analysis (FFT)
512x512pixels analysis
ANN classification
Featureextraction
2dim FFT
GLANDULAR
RESULTS: RESULTS: TEXTURE ANALYSISTEXTURE ANALYSIS
DENSE ADIPOSE GLANDULAR
DENSE >95% 0% 0%
ADIPOSE 16% 68±3% 16%
GLANDULAR 4% 3% 93±1%
SPICULATED LESIONS
Unroll spirals
Spatial frequencies analysis(FFT)
FF-ANN
examples
0 50 100 150 200 250 300
160
170
180
190
200
210
220 Vettore Spiral
f(j)
j
0 2 4 6 8 10 12 14 16
0
20
40
60
80
100 Trasformata di Fourier
F(k)
k
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
RESULTS @ sensitivity=90(±3)%:
Integration range 2-5
Spiculated lesions:CAD performances
Red= radiologist
Blue= CAD
RESULTS: RESULTS: SPICULATED LESIONSSPICULATED LESIONS
Sensitivity (per patient) 90±3%
FALSE POSITIVES / IMAGE 1.4
AVERAGE ROI 25 cm2
DATA REDUCTION ~ 10
Examples
MICROCALCIFICATION CLUSTERS
FF-ANN + Sanger learning rule
PCA
Method
•Image Preprocessing (convolution filters)
•PCA through a NN trained with the Sanger rule
•Study of the first Principal Components
•Classification
Preprocessing
• 60x60 pixels windows selection• convolution filters with dims:
5x5 7x7 9x9
Best results with a 7x7 filter with A=1\N2 aij <0 (aij kernel element)
Results
Sensitivity = 73 ± 2 % Specificity= 94 ± 2 %
With micro-calcification clusters
No Micro-calcification clusters
2
3
1
Micro-calcification clusters: CAD
Red= radiologist
Blue= CAD
RESULTS: RESULTS: MICRO-CALCIFICATION CLUSTERSMICRO-CALCIFICATION CLUSTERS
SENSITIVITY 73±2%
SPECIFICITY 94±2%
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