Computer-aided diagnosis

From Wikipedia, the free encyclopedia

Computer-Aided Diagnosis (CAD) is an inter-discpline combining computer science, especially artificial intelligence and digital image processing, and radiography for the automatic detection of certain diseases from medical images produced by various imaging modalties.

CAD finds most of its applications in the automatic detection of lesions in mammography. However, it is recently gaining more attention and will extend its applications into more branches of radiography.

A CAD system usually comprises of four procedures: preprocessing of images or enhancement; segmentation of regions of interests (from the possibly noisy background); feature selections of the regions of interests (ROI); and classifications of these features in the feature space, essentially a pattern recognition step.

Contents 1 Preprocessing 2 Segmentation 3 Feature selection 4 Classification 5 Criterion

[edit] Preprocessing

-Traditionally: gray scale conversion, histogram conversion, color composition, color conversion between RGB and HIS.
-Recent years: Wavelet based algorithms for enhancement

Procedure: Denoising, signal matching, sharpening

[edit] Segmentation

Purpose:Screening out the ROI
Edge-based:Object boundaries
Region-based:
Intensity of pixels
Advantage: brighter in ROI.

[edit] Feature selection

The following features have been reported in the literature:
-Areas of the ROIs;
-Compactness;
-Average gray level of the object
-Average gray level of border of the object
-Gradient strength of the object's perimeter
-Contrast: difference between the maximum gray level in the region and the mean of the border
-Maximum gray level inside the region
-Maximum edge gradient of the border.
-Variation in size
-Variation in pixel values
-Shape irregularity of the ROI

[edit] Classification

• Traditional:
  • KNN rule (CMEANS)
  • minimum distance classifier
  • Cascade Classifier
  • Bayesian classifier
• Artificial Neural Network (ANN):
  • Multilayer perceptron
  • Radial basis function network (RBF)
  • SVM

[edit] Criterion

• Receiver Operation Characteristics (ROC)