Biomedical Image Analysis and
Machine Learning
BMI 731 Winter 2005
Kun Huang
Department of Biomedical Informatics
Ohio State University

Introduction to biomedical imaging

Imaging modalities

Components of an imaging system

Areas of image analysis

Machine learning and image analysis

Why imaging?

Diagnosis
X

ray, MRI, Ultrasound, microscopic imaging (pathology
and histology) …

Visualization (invasive and noninvasive)
3

D, 4

D

Functional analysis
Functional MRI

Phenotyping
Microscopic imaging for different genotypes, molecular
imaging

Quantification
Cell count, volume rendering, Ca
2+
concentration …

Imaging modalities

Wavelength

Electron microscope

X

ray

UV

Light

Ultrasound

MRI

Fluorescence

Multi

spectral

Tomography

Video
Ultrasound

Components of Imaging System

Instrumentation :

Electrical engineering, physics, histochemistry …

Image generation

Sensor technology (e.g., scanner), coloring agents …

Image processing and enhancement

Both software, hardware, or experimental (dynamic
contrast)

Image analysis at all levels

Image processing, computer vision, machine learning

Manual/interactive

Image storage and retrieval

Database/data warehouse

Areas of Image Processing and Analysis

Image enhancement

Color correction, noise removal, contrast enhancement …

Feature extraction

color, point, edge (line, curves), area

cell, tissue type, organ, region

Segmentation

Registration

3

D reconstruction

Visualization

Quantization

Image Analysis and Machine Learning

Why machine learning

Classification at all levels

Pixel, texture, object …

Pattern recognition, statistical learning, multivariate
analysis …

Statistical properties
Curtersy of Raghu Machiraju

Common machine learning techniques

Dimensionality reduction

Principal component analysis (PCA, SVD, KLT)

Linear discriminant analysis (LDA, Fisher’s discriminant)
stack
PCA

Common machine learning techniques

Supervised learning
Learning
algorithm
Classifier
?

Neural network, Support vector machine (SVM),
MCMC, Bayesian network …

Common machine learning techniques

Unsupervised learning

K

means, K

subspaces, GPCA, hierarchical
clustering, vector quantization, …

Dimensionality Reduction

Principal component analysis (PCA)

Singular value decomposition (SVD)

Karhunen

Loeve
transform (KLT)
Basis for
P
SVD

Dimensionality Reduction

Principal component analysis (PCA)
=
=

Dimensionality Reduction

Principal component analysis (PCA)
=
≈
Knee point
Optimal in the sense of least square error.

Principal Component Analysis (PCA)

Geometric meaning

Fitting a low

dimensional linear model to data
Find
m
and
E
such that
J
is minimized.

Principal Component Analysis (PCA)

Statistical meaning

Direction with the largest variance

Principal Component Analysis (PCA)

Algebraic meaning

Energy

Principal Component Analysis (PCA)

Application : face recognition (Jon Krueger et. al.)
Average face
Eigenfaces
–
Principal Components

Linear Discriminant Analysis
B
.
2.0
1.5
1.0
0.5
0.5 1.0 1.5 2.0
.
.
.
.
.
.
.
.
.
.
.
.
.
A
w
.
(From S. Wu’s website)
Linear Discriminant Analysis
B
.
2.0
1.5
1.0
0.5
0.5 1.0 1.5 2.0
.
.
.
.
.
.
.
.
.
.
.
.
.
A
w
.
(From S. Wu’s website)

Linear Discriminant Analysis (PCA)

Which direction is a good one to pick?

Maximize the inter

cluster distance

Minimize the intra

cluster distance

Compromise : maximize the ratio between the
above two distances

Next time

Supervised learning

SVM

Unsupervised learning
–
K

means

Spectral clustering
OR

CT, Radon transform backprojection

MRI

Other image processing techniques (filtering,
convolution, color and contrast correction …)
Comments 0
Log in to post a comment