Transcript Simple and non-invasive Liver Fibrosis stage prediction method

Neural Network Ensemble based on Feature
Selection for Non-Invasive Recognition of
Liver Fibrosis Stage
Bartosz KRAWCZYK, Michał WOŹNIAK, Tomasz ORCZYK,
Piotr PORWIK, Joanna MUSIALIK, Barbara BŁOŃSKA-FAJFROWSKA
Presentation agenda



Overview.
Current diagnostic methods.
Proposed method.




Analyzed data.
Data analysis methods.
Result comparison.
Conclusions.
Overview

Liver fibrosis:

Accumulation of tough, fibrous scar tissue in the liver.
~1,75% of Poland’s population is infected with HCV.

Unthreated may cause Liver Cirrhosis and death.


Risk factors:




Chronic infection with hepatitis C or hepatitis B virus (HCV, HBV).
Immune system compromise (HIV or immunosuppressive drugs).
Heavy alcohol consumption.
Gradation indexes:



Knodell Histological Activity Index (HAI Score).
Ishak system.
METAVIR system.
Current diagnostic methods

Invasive

Liver biopsy




Risk of health complications or even death.
Up to 45% uncertainty depending on bioptate quality and size.
Still assumed as a „gold standard”.
Non-invasive


ELF Test
FibroTest & FibroScan


Expensive
Not very accurate
Proposed method

Non-invasive


Inexpensive


Blood test based
Only regular blood tests
Comparable with other non-invasive methods

Similar error level to FibroTest
Proposed method:
Analyzed data and problems

Data characteristics:




127 patients mostly with HCV (70%) and Liver Fibrosis.
All patients otherwise healthy and not under therapy.
34 parameters measured.
Problems:




Low data samples count.
Unequal distribution of diagnosed fibrosis level.
Incomplete records.
Many poor quality biopsies.
Proposed method:
Analyzed data and problems
Age* (years)
50 (13)
Male, n(%)
75 (59%)
Biopsy result, n(%)
F0
2 (2%)
F1
35 (28%)
F2
5 (4%)
F3
16 (13%)
F4
67 (53%)
HCV/HBV/other
70% / 9% / 21%
HB* (g/L)
RBC* (106/UL)
WBC* (103/UL)
PLT* (103/UL)
PT* (sec.)
PTP* (%)
APTT* (sec.)
INR*
ASPT* (IU/L)
ALAT* (IU/L)
ALP* (IU/L)
BIL* (mg/dL)
GGTP* (IU/L)
KREA* (mg/dL)
GLU* (mg/dL)
Na* (mmol/L)
14 (1.91)
5 (0.74)
6 (2.31)
161 (70.75)
13 (9.04)
90 (17.82)
38 (12.53)
1 (0.26)
65 (51.01)
72 (61.81)
104 (55.11)
2 (2.69)
89 (94.43)
1 (0.23)
95 (19.02)
138 (3.48)
K* (mmol/L)
Fe* (mmol/L)
CRP* (IU/L)
TG* (mg/dL)
CHO* (mg/dL)
Ur. acid* (mg/dL)
TP* (g/dL)
TIBC*
Neutr* (103/UL)
Lymph* (103/UL)
Mono* (103/UL)
Eos* (103/UL)
Baso* (103/UL)
Albu* (%)
Glb. α1* (%)
Glb. α2* (%)
Glb. β* (%)
Glb. γ* (%)
5 (5.16)
104 (70.23)
4 (25.38)
107 (50.83)
189 (51.04)
6 (1.39)
7 (0.81)
322 (120.47)
3 (1.35)
2 (0.55)
1 (0.19)
0 (0.13)
0 (0.02)
58 (7.79)
3 (1.33)
8 (2.52)
11 (2.43)
19 (7.21)
Proposed method:
Neural Network Ensemble
The introduced method of classifier ensemble design
consists of three main steps:

Building the pool of individual classifiers.

Pruning the acquired pool by discarding redundant
predictors.

Using a sophisticated trained fuser to deliver the
ensemble.
Proposed method:
Building the pool of classifiers



Models should be complementary to each other,
exhibiting at the same time high accuracy and high
diversity.
There is no single optimal approach for feature selection
task and results obtained on the basis of different
methods may differ significantly.
Instead of selecting a single best feature selection method
we use several of them to reduce the dimensionality of
the feature space.
Proposed method:
Ensemble pruning



There are several different ways in the literature on how
to select valuable members to the committee.
Ideal ensemble consists of classifiers of high individual
accuracy and high diversity.
Among diversity measures there are two major types:



Pairwise (shows how two classifiers differ from each other).
Non-pairwise (measure the diversity of the whole ensemble).
For measuring the diversity of whole ensemble we used
the entropy measure.
Proposed method:
Fusion of individual classifiers

Classifier fusion algorithms can make decisions on the
basis of class labels given by individual classifiers or they
can construct new discriminant functions on the basis of
individual classifier support functions:



The first group includes voting algorithms.
The second group is based on discriminant analysis.
The design of improved fusion classification models,
especially trained fusers, is the focus of current research.
Proposed method:
Fusion of individual classifiers
 , in,...,
Assume that we have K classifiers
apool after the
pruning procedure.
For a given object x  X each individual classifier decides for

i    1,...,Mbased
class
on the values of discriminants. Let
denote
that is assigned to class i for a given value of

F (l ) ia, xfunction
x, and that is used by the l-th classifier
. The
 (l )
combined
classifier uses the decision rule
 ( x)  i if Fˆ (i, x)  max Fˆ (k , x) , where Fˆ (i, x)    F (i, x) and    1.
The weights can be set dependent on the classifier and class
 (l ) (i)
number: weight
is assigned
to the l-th classifier and the ith class, and given classifier weights assigned to different
classes may differ.
(1)
K
l 1
(K )
K
(l )
kM
( 2)
(l )
(l )
i 1
Proposed method:
Feature selection algorithms
Eight different feature selection algorithms were used, namely:
 ReliefF,
 Fast Correlation Based Filter,
 Genetic Wrapper,
 Simulated Annealing Wrapper,
 Forward Selection,
 Backward Selection,
 Quick Branch & Bound,
 Las Vegas Incremental.
Neural network architecture was as follows: the number of neurons in the
input layer was equal to the number of selected features, the number of
output neurons was equal to the number of classes and the number of hidden
neurons was equal to the half of the sum of number of neurons from the
former layers.
Proposed method:
Set-up

As reference methods we have selected most popular
ensembles - Bagging, Boosting, Random Forest and
Random Subspace.

Additionally we have compared our method with the
single best classier from the pool, all classifiers from the
pool and with simple majority voting.

The combined 5x2 CV F test [1] was carried out to asses
the statistical signicance of obtained results.
Proposed method:
Results
Proposed method:
Results




The proposed neural network ensemble, based on feature
selection methods, has outperformed all the previously
used MCS for this problem.
The weakest results were returned by single best model
approach, which highlights the usefulness of utilizing more
than one classier to fully exploit the outputs of feature
selection methods.
Second biggest accuracy boost lies in the used fusertrained fusion of individual classiers allows to derive an
optimal linear combination of them.
The pruning step had smallest but still statistically
significant impact on the ensemble design.
Conclusions:



The presented paper shows that, despite some problems it is
possible to reach similar or even lower error level than
commercial tests.
It is also worth to mention that liver biopsy result, according
to the other research, is also only a prediction with
classification error varying from 35% up to 45% , depending on
the sample size and count.
we proved that each of the three steps embedded in the
proposed committee design has an important impact on the
quality of the final prediction and thus should not be omitted.
Thank you
for your attention
Contact:
[email protected]
[email protected]