Introduction to Medical Imaging

Mammography and Computer Aided Diagnostic (CAD) Example

Guy Gilboa

Course 046831

Mammography

Mammography purpose

Mammography is the process of using low-energy X-rays (usually around 30 kVp) to examine the human breast, which is used as a diagnostic and screening tool. The goal of mammography is the early detection of breast cancer, typically through detection of characteristic masses and/or microcalcifications.

Basic structure

Obtained images

CADComputer-Aided Detection and Diagnosis (CAD) of breast cancer with mammography Early detection is an effective way to diagnose and

manage breast cancer. Computer-aided detection or diagnosis (CAD) systems

can play a key role in the early detection of breast cancer and can reduce the death rate among women with breast cancer.

(*) Based on the paper J. Tang et al. "Computer-aided detection and diagnosis of breast cancer with mammography: recent advances." Information Technology in Biomedicine, IEEE Transactions on 13.2 (2009): 236-251.

Example – aiding radiologists

An original mammogram (left) and after MED-SEG processing (right), indicating a region of interest in white.

Taken from http://www.laserfocusworld.com/blogs/photon-focus/2011/10.html

Need for CAD – early detectionAt present, there are no effective ways

to prevent breast cancer, because its cause remains unknown.

However, efficient diagnosis of breast cancer in its early stages can give a woman a better chance of full recovery.

Therefore, early detection of breast cancer can play an important role in reducing the associated morbidity and mortality rates.

Need for CAD (cont’)Computer-aided detection or diagnosis

(CAD) systems use computer technologies to detect abnormalities in mammograms.

Abnormalities include calcifications, masses, and architectural distortion.

The use of these results by radiologists for diagnosis can play a key role in the early detection of breast cancer and help to reduce the death rate among women with breast cancer.

Medical background - Microcalcifications (MC)MCs are tiny deposits of calcium that appear

as small bright spots in mammograms. Clustered MCs can be an important indicator

of breast cancer. They appear in 30%–50% of cases

diagnosed by mammographic screenings

MC detection methods

4 broad categories: 1. Basic image enhancement.2. Stochastic modeling.3. Multiscale decomposition.4. Machine learning methods.

Image enhancementMotivation: MCs tend to be brighter

than their surroundings. Image enhancement methods are used

to improve the contrast of MCs, and then apply a threshold to separate them from their surroundings.

Denoising Thresholding

Post-processing

(e.g. morphological operators)

Stochastic Modeling Utilize statistical differences between MCs and their

surroundings. For instance, Gurcan et al. used differences in higher

order statistics [e.g., the third moment (skewness) and the fourth moment (kurtosis)], where it was conjectured that areas with no MCs would have a Gaussian-like distribution and areas with MCs would be non-Gaussian (nonzero skewness and kurtosis).

This approach can be prone to errors in background (non-MC) regions that are spatially variant.

Advanced methods today use Markov random field (MRF) and Gaussian mixture models.

However, estimating a proper prior distribution remains a challenging task in these probabilistic approaches.

Multiscale Decomposition Exploits the differences in frequency content between

MC spots and their surrounding background. In particular, wavelet transforms have been widely

investigated for MC detection. For instance, Strickland and Hahn used undecimated

biorthognal wavelet transforms in which MCs were represented by circular Gaussian shapes with varying widths along the different scales. The undecimated wavelet transform has the advantage of being translation invariant. Optimal subband weighting was applied prior to reconstruction from subbands 2 and 3 for improved detection and segmentation of clustered MCs.

These methods are often used as feature extraction techniques that are used in conjunction with other approaches (e.g., as input to classifiers).

Machine Learning methods Aim to decipher dependencies from the data. In the context of MC detection, the problem is typically

treated as a binary classification process, where the goal is to determine whether an MC is present or not at a pixel location.

As an example, Yu and Guan proposed a two-stage neural network approach, where wavelet components, gray-level statistics, and shape features were used to train a two-stage network. ◦ The first stage identifies potential MC pixels in the mammograms.◦ The second stage detects individual MC objects.

Machine learning methods have received the largest share of research in recent developments.

Machine-learning-based methods seemed to have achieved the best performance.

Example – using wavelets

Jinghuan, Guo, et al. "Study on Microcalcification Detection Using Wavelet Singularity." International Journal of Signal Processing, Image Processing & Pattern Recognition 7.1 (2014).

MC detection summary

Taken from (*)

Sensitivity-Specificity curves

T=True; F=False; P=Positive; N=NegativeSensitivity – (true positive)/(total ill

cases)=TP/(TP+FN)Specificity – (true negative)/(total well

cases) = TN/(TN+FP)

CAD generic procedure

Medical image

Other CAD examples - retina

Detection of colon polyps

From http://www.med.nagoya-u.ac.jp/english01/6489/6600/Diagnostic-Medical-Image-Processing.html

Cardiac – stenosis assessment

The investigational CAD algorithm automatically tracks and segments the coronary tree, extracting vessel centerlines and labeling them before performing automated stenosis detection and marking of target vessels.

From http://www.auntminnie.com/index.aspx?sec=ser&sub=def&pag=dis&ItemID=90789

Exam exercise Reminder FDG – a PET

radiotracerQ.3 Moed A, Q.7 Moed A

What have we learnt in this course?

The main modalities X-ray, CT, MRI, Ultrasound, PET:

◦The basic physics behind them. ◦The main medical applications.

Image processing of medical data◦Tomography and reconstruction◦ Denoising◦Segmentation◦Motion compensation and deconvolution◦RegistrationI hope you enjoyed,

Good luck!