Mammographic image analysis - Iac-Cnr

Bayesian analysis of dynamic
MR breast images
P. Barone, F. de Pasquale, G. Sebastiani
Istituto per le Applicazioni del Calcolo ‘‘M. Picone’’
CNR, Rome (Italy)
J. Stander
Department of Mathematics and Statistics
University of Plymouth, UK
M. Crecco, A. M. Di Nallo, F. P. Gentile
‘‘Istituto Regina Elena’’
Rome (Italy)
1
Introduction
Data and aims of the analysis:
• Dynamic breast MRI consists of a temporal sequence of
images of the same slice acquired after the injection of a
contrast agent into the blood stream
• Typically, for breast studies, a few tens of 256 x 256
images are acquired consecutively
• ‘In vivo’ tissue perfusion maps in an organ of interest
(breast)
2
• These data are typically affected by two main kind of
degradation:
random degradation due to the measurement noise;
deterministic degradation due to the patient motion
• Radiologists’s aim is to extract as much clinical
information as possible from the image sequence as few
images of easy interpretation and with low degradation
(label each pixel of the image as non tumoral, benign
tumoral and malign tumoral tissues)
• ‘real time’ analysis
3
Data
first sequence image
last sequence image
4
Non-parametric approach
• Bayesian estimation of true image intensity for each pixel
of a selected R.O.I. and for each time without using a
parametric spatio temporal model for the acquired signal
• Relevant quantities (class attributes) describing image
intensity variation are computed independently for each
pixel from these estimated image intensity profiles
• These are used to classify the underlying scene into a
certain number of categories based on the temporal pattern
of intensity
5
Estimation results
Original R.O.I
Estimated R.O.I.
Patient 1
Patient 2
Patient 3
6
Non-parametric approach:
estimated attributes
First attribute
Second attribute
7
Parametric approach
• We adopted here a parametric spatio temporal model for
the acquired signal to describe true image intensities at
each pixel
• Model parameters are then displayed as images and can be
used for image classification
• After paramater estimation, the model can be used to
estimate true image intensities.
8
Spatio temporal model paramaters
9
Classification
The data in the classifica tion procedure are given by :
d i  d1,i ,..., d m,i ,
i  1,...,n
where n is the number of pixels analysed and m is the number
of attributes available for each pixel (in our study 2)
Pixel i  k i  1,2,..., K 
K classes .
we consider K  3
10
Classification results
11