Diagnosis%20of%20Ovarian%20Cancer%20Based%20on%20Mass%20Spectra%20of%20Blood%20Samples

1
Diagnosis of Ovarian CancerBased on Mass
Spectraof Blood Samples
Hong Tang Yelena Mukomel Eugene Fink
2
Motivation
Early detection of cancer byanalysis of blood
samples.

• Fast inexpensive test
• Little discomfort

3
Outline

• Mass-spectrum curves
• Feature extraction
• Experimental results
• Conclusions

4
Mass spectrum
102
100
signalintensity
102
104
5,000
10,000
15,000
20,000
0
ratio of molecular weightto net electric charge
The curve of a cancer patient usuallydiffers
from that of a healthy person.
5
Patient data

• Mass-spectrum curves of 685 people
• Every curve consists of 15,155 points

Data set Number of cases Number of cases
Data set Cancer Healthy
1 2 3 100 100 162 116 116 91
6
Outline

• Mass-spectrum curves
• Feature extraction
• Experimental results
• Conclusions

7
Candidate features

• Every point of the mass-spectrum curve is a
  candidate feature

• Its relevance depends on the mean difference
  between values for cancer patients and healthy
  people

8
Feature relevance

• Mean difference ?c ?h

• Standard deviation of the difference (?c2
  ?h2)0.5

signal intensity
9
Minimal distance
102
100
signalintensity
102
feature
104
min distance

• Impose a lower bound on the distance between
  feature points, which prevents the selection
  of correlated features

• After selecting a feature point, discard all
  points within this distance bound

10
Feature selection
102
100
signalintensity
102
104

• Repeat for a given number of features
• Select the most relevant feature point
• Discard all points within the minimal
  distance from the selected feature

11
Outline

• Mass-spectrum curves
• Feature extraction
• Experimental results
• Conclusions

12
Control variables

• Number of feature points 1 to 64

• Min distance between features 1 to 1024

• Data mining techniques Decision trees
  (C4.5) Support vector machines (SVMFu)
  Neural networks (Cascor 1.2)

13
Measurements

• Sensitivity Probability of the correct
  diagnosis for a cancer patient

• Specificity Probability of the correct
  diagnosis for a healthy person

14
Results
Num. offeatures Min.dist. Sensi-tivity Speci-ficity
Set 1 DTSVMNN 43232 1 16256 86 82 80 78 84 84
Set 2 DTSVMNN 8 432 4 2 1 92 96 93 96 93 98
Set 3 DTSVMNN 81616 64 8 2 98100100 100 99 99
15
Summary

• Performance range
• Sensitivity 80100
• Specificity 78100

16
Summary
Performance range
Sensitivity Specificity
Set 1 8086 7884
Set 2 9296 9396
Set 3 98100 99100

• Optimal parameters
• Number of feature points 432
• Min distances between features 1256
• Data mining technique Any

17
Outline

• Mass-spectrum curves
• Feature extraction
• Experimental results
• Conclusions

18
Conclusions
We have developed a technique for thedetection
of ovarian cancer based on theanalysis of blood
mass spectra.
The accuracy of this technique is still low, and
results vary across data sets.
19
Future work

• Use more patient data

• Consider other features of mass-spectrum
  curves

• Apply to other cancers