View Learning: An extension to SRL An application in Mammography

1
View Learning An extension to SRL An application
in Mammography

• Jesse Davis, Beth Burnside, Inês Dutra Vítor
  Santos Costa, David Page, Jude Shavlik Raghu
  Ramakrishnan

2
Background

• Breast cancer is the most common cancer
• Mammography is the only proven screening test
• At this time approximately 61 of women have had
  a mammogram in the last 2 years
• Translates into 20 million mammograms per year

3
The Problem

• Radiologists interpret mammograms
• Variability in among radiologists
• differences in training and experience
• Experts have higher cancer detection and less
  benign biopsies
• Shortage of experts

4
Common Mammography findings

• Microcalcifications
• Masses
• Architectural distortion

5
Calcifications
6
Mass
7
Architectural distortion
8
Other important features

• Microcalcifications
• Shape, distribution, stability
• Masses
• Shape, margin, density, size, stability
• Associated findings
• Breast Density

9
Other variables influence risk

• Demographic risk factors
• Family History
• Hormone therapy
• Age

10
Standardization of Practice

• -Passage of the Mammography Quality Standards Act
  (MQSA) in 1992
• -Requires tracking of patient outcomes through
  regular audits of mammography interpretations and
  cases of breast cancer
• -Standardized lexicon BI-RADS was developed
  incorporating 5 categories that include 43 unique
  descriptors

11
BI-RADS
Margins -circumscribed -microlobulated -obscured -
indistinct -Spiculated
Special Cases
Mass
Associated Findings
Skin Thickening
Shape -round -oval -lobular -irregular
Tubular Density
Density -high -equal -low -fat containing
Skin Lesion
Lymph Node
Architectural Distortion
Trabecular Thickening
Assymetric Breast Tissue
Calcifications
Typically Benign -skin -vascular -coarse/popcorn -
rod-like -round -lucent-centered -eggshell/rim -mi
lk of calcium -suture -dystrophic -punctate
Nipple Retraction
Distribution -clustered -linear -segmental -region
al -diffuse/scattered
Focal Assymetric Density
Intermediate -amorphous
Axillary Adenopathy
Skin Retraction
Higher Probability Malignancy -pleomorphic -fine/l
inear/branching
12
Mammography Database

• Radiologist interpretation of mammogram
• Patient may have multiple mammograms
• A mammogram may have multiple abnormalities
• Expert defined Bayes net for determining whether
  an abnormality is malignant

13
Original Expert Structure
14
Mammography Database
15
Types of Learning

• Hierarchy of types of learning that we can
  perform on the Mammography database

16
Level 1 Parameters
Given Features (node labels, or fields in
database), Data, Bayes net structure Learn
Probabilities. Note probabilities needed are
Pr(Be/Mal), Pr(ShapeBe/Mal), Pr (SizeBe/Mal)
17
Level 2 Structure
Be/Mal
Given Features, Data Learn Bayes
net structure and probabilities. Note with this
structure, now will need Pr(SizeShape,Be/Mal)
instead of Pr(SizeBe/Mal).
Shape
Size
18
Mammography Database
19
Mammography Database
20
Mammography Database
21
Level 3 Aggregates
Given Features, Data, Background knowledge
aggregation functions such as average, mode, max,
etc. Learn Useful aggregate features,
Bayes net structure that uses these features, and
probabilities. New features may use other
rows/tables.
Avg size this date
Be/Mal
Shape
Size
22
Mammography Database
23
Mammography Database
24
Mammography Database
25
Level 4 View Learning
Given Features, Data, Background knowledge
aggregation functions and intensionally-defined
relations such as increase or same
location Learn Useful new features defined by
views (equivalent to rules or SQL queries), Bayes
net structure, and probabilities.
Shape change in abnormality at this location
Increase in average size of abnormalities
Avg size this date
Be/Mal
Shape
Size
26
Structure Learning Algorithms

• Three different algorithms
• Naïve Bayes
• Tree Augmented Naïve Bayes (TAN)
• Sparse Candidate Algorithm

27
Naïve Bayes Net

• Simple, computationally efficient

28
Example TAN Net

• Also computationally efficient
• Friedman,Geiger Goldszmidt 97

Class Value

29
TAN

• Arc from class variable to each attribute
• Less Restrictive than Naïve Bayes
• Each attribute permitted at most one extra parent
• Polynomial time bound on constructing network
• O(( attributes)2 training set)
• Guaranteed to maximize LL(BT D)

30
TAN Algorithm

• Constructs a complete graph between all the
  attributes (excluding class variable)
• Edge weight is conditional mutual information
  between the vertices
• Find maximum weight spanning tree over the graph
• Pick root in tree and make edges directed
• Add edges from directed tree to network

31
General Bayes Net
32
Sparse Candidate

• Friedman et al 97
• No restrictions on directionality of arcs for
  class attribute
• Limits possible parents for each node to a small
  candidate set

33
Sparse Candidate Algorithm

• Greedy hill climbing search with restarts
• Initial structure is empty graph
• Score graph using BDe metric (Cooper Herskovits
  92, Heckerman 96)
• Selects candidate set using an information metric
• Re-estimate candidate set after each restart

34
Sparse Candidate Algorithm

• We looked at several initial structures
• Expert structure
• Naïve Bayes
• TAN
• Scored network on tune set accuracy

35
Our Initial Approach for Level 4

• Use ILP to learn rules predictive of malignant
• Treat the rules as intensional definitions of new
  fields
• The new view consists of the original table
  extended with the new fields

36
Using Views

• malignant(A) -
• massesStability(A,increasing),
• prior_mammogram(A,B,_),
• H0_BreastCA(B,hxDCorLC).

37
Sample Rule

• malignant(A) -
• BIRADS_category(A,b5),
• MassPAO(A,present),
• MassesDensity'(A,high),
• HO_BreastCA(A,hxDCorLC),
• in_same_mammogram(A,B),
• Calc_Pleomorphic(B,notPresent),
• Calc_Punctate(B,notPresent).

38
Methodology

• 10 fold cross validation
• Split at the patient level
• Roughly 40 malignant cases and 6000 benign cases
  in each fold

39
Methodology

• Without the ILP rules
• 6 folds for training set
• 3 folds for tuning set
• With ILP
• 4 folds to learn ILP rules
• 3 folds for training set
• 2 folds for tuning set
• TAN/Naïve Bayes dont require tune set

40
Evaluation

• Precision and recall curves
• Why not ROC curves?
• With many negatives ROC curves look overly
  optimistic
• Large change in number of false positives yields
  small change in ROC curve
• Pooled results over all 10 folds

41
ROC Level 2 (TAN) vs. Level 1
42
Precision-Recall Curves
43
(No Transcript)
44
(No Transcript)
45
Related Work ILP for Feature Construction

• Pompe Kononenko, ILP95
• Srinivasan King, ILP97
• Perlich Provost, KDD03
• Neville, Jensen, Friedland and Hay, KDD03

46
Ways to Improve Performance

• Learn rules to predict benign as well as
  malignant.
• Use Gleaner (Goadrich, Oliphant Shavlik,
  ILP04) to get better spread of Precision vs.
  Recall in the learned rules.
• Incorporate aggregation into the ILP runs
  themselves.

47
Richer View Learning Approaches

• Learn rules predictive of other fields.
• Use WARMR or other first-order clustering
  approaches.
• Integrate Structure Learning and View
  Learningscore a rule by how much it helps the
  current model when added

48
Level 4 View Learning
Given Features, Data, Background knowledge
aggregation functions and intensionally-defined
relations such as increase or same
location Learn Useful new features defined by
views (equivalent to rules or SQL queries), Bayes
net structure, and probabilities.
Shape change in abnormality at this location
Increase in average size of abnormalities
Avg size this date
Be/Mal
Shape
Size
49
Integrated View/Structure Learning
sc(X)- id(X,P), id(Y,P), loc(X,L), loc(Y,L),
date(Y,D1), date(X,D2), before(D1,D2),
shape(X,Sh1), shape(Y,Sh2), Sh1 \ Sh2.
Increase in average size of abnormalities
Avg size this date
Be/Mal
Shape
Size
50
Integrated View/Structure Learning
sc(X)- id(X,P), id(Y,P), loc(X,L), loc(Y,L),
date(Y,D1), date(X,D2), before(D1,D2),
shape(X,Sh1), shape(Y,Sh2), Sh1 \ Sh2,
size(X,S1), size(Y,S2), S1 gt S2.
Increase in average size of abnormalities
Avg size this date
Be/Mal
Shape
Size
51
Integrated View/Structure Learning
sc(X)- id(X,P), id(Y,P), loc(X,L), loc(Y,L),
date(Y,D1), date(X,D2), before(D1,D2),
shape(X,Sh1), shape(Y,Sh2), Sh1 \ Sh2,
size(X,S1), size(Y,S2), S1 gt S2.
Increase in average size of abnormalities
Avg size this date
Be/Mal
Shape
Size
52
Integrated View/Structure Learning
sc(X)- id(X,P), id(Y,P), loc(X,L), loc(Y,L),
date(Y,D1), date(X,D2), before(D1,D2),
shape(X,Sh1), shape(Y,Sh2), Sh1 \ Sh2,
size(X,S1), size(Y,S2), S1 gt S2.
Avg size this date
Be/Mal
Shape
Size
53
Richer View Learning (Cont.)

• Learning new tables
• Just rules for non-unary predicates
• Train on pairs of malignancies for the same
  mammogram or patient
• Train on pairs (triples, etc.) of fields, where
  pairs of values that appear in rows for malignant
  abnormalities are positive examples, while those
  that appear only in rows for benign are negative
  examples

54
Conclusions

• Graphical models over databases were originally
  limited to the schema provided
• Humans find it useful to define new views of a
  database (new fields or tables intensionally
  defined from existing data)
• View learning appears to have promise for
  increasing the capabilities of graphical models
  over relational databases, perhaps other SRL
  approaches

55
WILD Group

• Jesse Davis
• Beth Burnside
• Ines Dutra
• Vitor Santos Costa
• Raghu Ramakrishnan
• Jude Shavlik
• David Page

• Others
• Hector Corrada-Bravo
• Irene Ong
• Mark Goadrich
• Louis Oliphant
• Bee-Chung Chen