Diagnosis of Ovarian Cancer Based on Mass Spectrum of Blood Samples Committee: Eugene Fink Lihua Li Dmitry B. Goldgof Hong Tang
Diagnosis of Ovarian Cancer Based on Mass Spectrum of Blood Samples
Feb 25, 2016
Diagnosis of Ovarian Cancer Based on Mass Spectrum of Blood Samples. Hong Tang. Committee: Eugene Fink Lihua Li Dmitry B. Goldgof. Outline. Introduction Previous work Feature selection Experiments. Motivation. Early cancer detection is critical for successful treatment. - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Diagnosis of Ovarian Cancer Based on Mass Spectrum of Blood Samples
Committee:Eugene Fink
Lihua LiDmitry B. Goldgof
Hong Tang
Outline
• Introduction
• Previous work
• Feature selection
• Experiments
Motivation
Early cancer detection is criticalfor successful treatment.
Five year survival for ovarian cancer:• Early stage: 90%• Late stage: 35%
80% are diagnosed at a late stage.
Motivation
Desired features ofcancer detection:
• Early detection
• High accuracy
• Low cost
Mass spectrum
We can detect some early-stage cancersby analyzing the blood mass spectrum.
ratio of molecular weight to electrical charge
inte
nsity
20,0000 5,000 10,000 15,000
10–4
10–2
100
102
Mass spectrumMass spectrum
Data miningResults
Blood
Outline
• Introduction
• Previous work
• Feature selection
• Experiments
Initial work
• Vlahou et al. (2001): Manual diagnosis
of bladder cancer based on mass spectra
• Petricoin et al. (2002): Application of
clustering to mass spectra for the ovarian-
cancer diagnosis
Decision treesAdam et al. (2002): 96% accuracy for prostate cancerQu et al. (2002): 98% accuracy for prostate cancer
Later work
Neural networksPoon et al. (2003): 91% accuracy for liver cancer
ClusteringPetricoin et al. (2002): 80% accuracy for prostate cancer
Outline
• Introduction
• Previous work
• Feature selection
• Experiments
Feature selection
ratio of molecular weight to electrical charge
inte
nsity
200 400 600
CancerHealthy
2 21 2 1 2/ Statistical difference:
Feature selection
ratio of molecular weight to electrical charge
inte
nsity
200 400 600
Window size: minimal distance between selected points
CancerHealthy
Outline
• Introduction
• Previous work
• Feature selection
• Experiments
Data sets
Dataset
Number of cases Cancer Healthy
123
100100162
116116 91
Learning algorithms
• Decision trees (C4.5)
• Support vector machines (SVMFu)
• Neural networks (Cascor 1.2)
Control variables
• Number of features, 1–64
• Window size, 1–1024
Best control valuesDecision trees
Data set
Number of features
Window size
Accuracy
1 4 1 82%2 8 4 94% 3 8 64 99%
Best control valuesSupport vector machines
Data set
Number of features
Window size
Accuracy
1 32 16 83%2 4 2 94% 3 16 8 99%
Best control valuesNeural networks
Data set
Number of features
Window size
Accuracy
1 32 256 82%2 32 1 96% 3 16 2 99%
Learning curveData set 1
accu
racy
(%)
training size
90
80
60
100
70
Decision trees, SVM, Neural networks
50 100 150 200 250
accu
racy
(%)
Learning curveData set 2
training size
90
80
60
100
70
Decision trees, SVM, Neural networks
0 50 100 150 200 250
Learning curveData set 3
accu
racy
(%)
training size
50 100 150 20060
70
90
80
100
0
Decision trees, SVM, Neural networks
250
Main results
Automated detection of ovarian cancer byanalyzing the mass spectrum of the blood
• Experimental comparison of decision
trees, SVM and neural networks
• Identification of the most informative
points of the mass-spectrum curves
Future work
• Experiments with other data sets
• Other methods for feature selection
• Combining with genetic algorithm
Related Documents
