Overview of Advanced Computer Vision

jabgoldfishAI and Robotics

Oct 19, 2013 (4 years and 2 months ago)

70 views

Overview of Advanced Computer Vision
Systems

for Skin Lesions Characterization

IEEE TRANSACTIONS ON INFORMATION
TECHNOLOGY IN BIOMEDICINE, VOL. 13, NO. 5,
SEPTEMBER 2009

Ilias Maglogiannis
, Member, IEEE
, and Charalampos N.
Doukas
, Student Member, IEEE

Presentor:
陳麒文

Outline


Skin cancer back
-

Ground information


Materials and methods


Image Acquisition Techniques


Definition of Features for the Classification of
Skin Lesions


Skin lesion classification methods


Results




Definition of Features for the Classification of
Skin Lesions


ABCD Rule


pattern analysis


Menzies method


seven
-
point checklist;


texture analysis

ABCD rules






asymmetry


border


color


differential structures


Pattern analysis


Menzies method


Seven
-
point check list


atypical pigment network, blue
-
whitish veil,
atypical vascular pattern


irregular streaks, irregular dots/globules, irregular
blotches, and regression structures


Texture analysis

SKIN LESION CLASSIFICATION
METHODS


Learning Phase


statistical


Neural networks


support vector machine (SVM)


adaptive wavelet
-
transform
-
based tree
-
structure
classification (ADWAT)


Testing Phase


Feature selection


The success of image recognition depends on
the correct selection of the features =>
optimization problem


heuristic strategies, greedy or genetic
algorithms


strategies from statistical pattern recognition



XVAL, LOO, SFFS, SBFS, PCA, GSFS

RESULTS FROM EXISTING
SYSTEMS


Conclusion


It is often difficult to differentiate early melanoma
from other benign skin lesions even for experienced


It is even more difficult for primary care physicians
and general practitioners


The early diagnosis of skin cancer is important for the
therapeutic procedure and reducing mortality rates.


Most remarkable features have been surveyed in this
paper


Cost of a simple CDSS for skin assessment is low


Standardization of all steps in the CDSS procedure
from the image acquisition until the feature extraction
and the classification stages is considered essential

Q&A