prognostic modelling and profiling of breast cancer patients after surgery n.
Download
Skip this Video
Loading SlideShow in 5 Seconds..
Prognostic Modelling and Profiling of Breast Cancer Patients after Surgery PowerPoint Presentation
Download Presentation
Prognostic Modelling and Profiling of Breast Cancer Patients after Surgery

Loading in 2 Seconds...

play fullscreen
1 / 20

Prognostic Modelling and Profiling of Breast Cancer Patients after Surgery - PowerPoint PPT Presentation


  • 111 Views
  • Uploaded on

Prognostic Modelling and Profiling of Breast Cancer Patients after Surgery. Ian Jarman School of Computer and Mathematical Sciences Liverpool John Moores University Supervisor: Prof. Paulo Lisboa. Contents. Motivation Background Prognostic Modelling Rule Extraction Summary Further Work.

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

PowerPoint Slideshow about 'Prognostic Modelling and Profiling of Breast Cancer Patients after Surgery' - diata


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
prognostic modelling and profiling of breast cancer patients after surgery

Prognostic Modelling and Profiling of Breast Cancer Patients after Surgery

Ian Jarman

School of Computer and Mathematical Sciences

Liverpool John Moores University

Supervisor: Prof. Paulo Lisboa

contents
Contents
  • Motivation
  • Background
  • Prognostic Modelling
  • Rule Extraction
  • Summary
  • Further Work
motivation
Motivation
  • Present models developed over 20 years ago
  • Introduction of Breast Screening
  • Increasing research into Artificial Neural Networks (ANN) for censored data
  • Add to the toolkit of the oncologist in support of their decisions
background
Background
  • Survival Analysis
  • Current Models
  • Artificial Neural Networks
  • Unlock the Black Box
    • Rule Extraction
survival analysis
Survival Analysis
  • Survivor Function [S(t)]
  • Hazard Function [H(t)]
    • instantaneous potential per unit time for the event to occur, given that the individual has survived to time t
  • Censored Data
    • When an individual drops out of a study for reasons other than the event of interest
current models
Current Models
  • Cox Proportional Hazard Model
    • Non parametric
      • no assumptions about the form of the data distribution
    • Linear in the parameters
  • Nottingham Prognostic Index (NPI) (0.2size + grade + nodal stage. )
artificial neural networks

Sigmoid Activation function

Such as:

1/ (1+ exp(-a))

weights

weights

bias

bias

hidden nodes

input

output

Artificial Neural Networks
  • Multi-Layer Perceptron (MLP)
    • Extension of logistic regression
artificial neural networks1
Artificial Neural Networks
  • PLANN-ARD

Partial Logistic Artificial Neural Network- Automatic Relevance Determination

    • Bayesian framework for network regularisation
    • Makes use of Censored Data
    • Irrelevant variables are‘soft pruned’
rule extraction osre
Rule Extraction (OSRE)
  • Developed by
    • Dr Terence Etchells
    • Prof. Paulo Lisboa
  • Finds explicit rules
    • e.g. patient is in a High Risk category if:
      • Nodes Ratio > 60% and Age between 40-59
prognostic modelling
Prognostic Modelling
  • NPI vs PLANN-ARD
  • Kaplan- Meier survival curves
cross tabulation matrix

Cox Lowest Risk

Highest Risk

PLANN Highest

Risk

Lowest Risk

Cross-tabulation Matrix
  • How well are the models correlated?
km survival within matrix

P

L

A

N

N

4

3

2

1

NIL

100% censored

n=1

100% censored

n=8

100% censored

n=41

100% censored

n=19

100% censored

n=35

NIL

NPI 1 2 3 4

KM Survival within Matrix
development of a new prognostic model

NIL

100% censored

n=1

100% censored

n=8

100% censored

n=41

100% censored

n=19

100% censored

n=35

NIL

4

3

2

1

Development of a New Prognostic Model
  • Group patients by survival
  • Distinct pattern emerges
how does survival differ
How Does Survival differ?
  • Statistically there is no difference!

Model by NPI

Model by PLANN-ARD

Model by new method

why continue

150

287

89

33

Why Continue?
  • Statistically the same, but patient grouping differs
rule extraction
Rule Extraction
  • Problem
    • Many rules can be produced to describe a data set
  • Solution
    • Develop a new methodology to refine the rules
boxed rules

++ ++ ++

++ ++ ++

++ ++ ++

++ ++ ++

+++ + + +

+ + + +++ ++

+ + + + + + + +++ + + +

+ + ++

+ + + + + +

+++ + +

+ + + +

+ ++ + + + + +

+ + + +

+ + + +

+ + +

+ + + +

+++++

+ + + + +

++ ++ ++++

+ + + ++ +

+ + + +

++++ ++

+ + +

++ ++ ++

++ ++ ++

++ ++ ++

++ ++ ++

+ + +

+

+

+

+

+ +

+

+ +

+

+

+

+

++ ++ ++

++ ++ ++

++ ++ ++

++ ++ ++

+ +

+

+ +

+

+

+ + + + +

+ + +++

++ + + ++

+ +

+ + +

+ ++

+ ++

++ +

+ + + + + + + +

+ + + + + + + +

+ + + + + + + +

++++++

+

++ ++ +

++

+ + + + + + + +

+ + + + + + + +

+ + + + + + + +

Rule Extraction

Decision Tree

Boxed Rules
roc curve

Acceptable specificity

ROC Curve
  • True Positives
    • Sensitivity
  • False Positives
    • 1-specificity
  • [1-specificity, sensitivity]
  • Refine Rules
summary
Summary
  • An analysis of new methods overdue
  • Development of New Prognostic Model
  • Prognostic Models
    • Statistically the same, but patient grouping differs
  • Rule Reduction Method for Rule Extraction
further work
Further Work
  • Use these methods for analysis of data
    • For one centre
    • Between centres
  • Visualisation techniques
    • ART, SOM