BMC Medical Informatics and Decision Making

Research

AdaBoost-based multiple SVM-RFE for classification of mammograms in DDSM

Sejong Yoon and Saejoon Kim*

Address: Department of Computer Science and Engineering, Sogang University,1 Shinsu-dong, Mapo-gu, Seoul, Korea Email: Sejong Yoon - sjyoon@sogang.ac.kr; Saejoon Kim* - saejoon@sogang.ac.kr * Corresponding author

from 2008 International Workshop on Biomedical and Health Informatics in conjunction with 2008 IEEE Conference of Bioinformatics and Biomedicine (BIBM)

Philadelphia, PA, USA. 3 November 2008 Published: 3 November 2009

BMC Medical Informatics and Decision Making 2009, 9(Suppl l):SI doi:l0.ll86/l472-6947-9-SI-SI

This article is available from: http://www.biomedcentral.com/l472-6947/9/Sl/Sl © 2009 Yoon and Kim; licensee BioMed Central Ltd.

This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Background: Digital mammography is one of the most promising options to diagnose breast cancer which is the most common cancer in women. However, its effectiveness is enfeebled due to the difficulty in distinguishing actual cancer lesions from benign abnormalities, which results in unnecessary biopsy referrals. To overcome this issue, computer aided diagnosis (CADx) using machine learning techniques have been studied worldwide. Since this is a classification problem and the number of features obtainable from a mammogram image is infinite, a feature selection method that is tailored for use in the CADx systems is needed.

Methods: We propose a feature selection method based on multiple support vector machine recursive feature elimination (MSVM-RFE). We compared our method with four previously proposed feature selection methods which use support vector machine as the base classifier. Experiments were performed on lesions extracted from the Digital Database of Screening Mammography, the largest public digital mammography database available. We measured average accuracy over 5-fold cross validation on the 8 datasets we extracted.

Results: Selecting from 8 features, conventional algorithms like SVM-RFE and multiple SVM-RFE showed slightly better performance than others. However, when selecting from 22 features, our proposed modified multiple SVM-RFE using boosting outperformed or was at least competitive to all others.

Conclusion: Our modified method may be a possible alternative to SVM-RFE or the original MSVM-RFE in many cases of interest. In the future, we need a specific method to effectively combine models trained during the feature selection process and a way to combine feature subsets generated from individual SVM-RFE instances.

Abstract

Background

Applications of artificial intelligence and machine learning techniques in medicine are now common and computer aided diagnosis (CADx) systems are one of those successful applications. Breast cancer, the most common cancer in women and second largest cause of death [1], is the disease which CADx systems are expected to be employed most successfully. To apply CADx systems, various imaging methods are available to reflect the inside tissue structure of breasts. Digital mammography using low-dose x-ray is one of those methods and is the most popular one worldwide. It has advantages over other methods such as sonar or magnetic resonance imaging (MRI) due to low cost and wide availability [2]. With digital mammography devices, doctors are able to find abnormal lesions which cannot be recognized using clinical palpation on breasts. CADx systems are applied on those images to detect and diagnose abnormalities. Since the early detection of breast cancer is important to ensure successful treatment of the disease, recent advances in research community have concentrated on improving the performance of CADx systems. Improvements in CADx systems can be obtained by solving two classification tasks: (1) detect more abnormalities or (2) distinguish actual malignant cancers from benign ones. Detecting abnormalities from a digitized mammogram is a relatively easy task and many improvements have been achieved while the latter is still a major area of research [3]. To achieve better performance, both classic and modern machine learning approaches such as Bayesian networks [4], artificial neural networks [5,6] and support vector machines (SVMs) [5,7] have been applied. However, the performance of CADx systems is still not as high as required for practical usage. This problem can be partially solved by using a better feature selection method that optimally fits to the mammogram classification problem [3].

We propose a new feature selection method for SVMs in this paper. Our method is based on SVM-Recursive Feature Elimination (SVM-RFE) [8] and its ensemble variant Multiple SVM-RFE [9]. We have conducted a comparison of the classification performance with baseline methods and two other SVM-RFE based feature selection methods, JOIN and ENSEMBLE, proposed by other groups [10]. To compare performances of methods, we prepared a dataset consisting of mass and calcification lesions extracted from Digital Database of Screening Mammography (DDSM) [11], the largest publicly available mammogram database.

Methods Notations

Let us suppose that a data set consists of N examples x1,..., xNeach of which has P features {1,..., P}.

Let xn = (x1; n,..., xP, n) be the n-th example where n e {1,..., N}, and the i-th feature value, i e {1,..., P}, of the n-th example is denoted by xi, n. Class labels of the N examples will be denoted by y = (y1,..., yN).

In this paper, we only consider a binary classification problem because we are interested in distinguishing benign and malignant examples. Overall, the labeled data set is expressed as {(x1, y1),..., (xN, yN)}.

SVM is one of the most popular modern classification methods. Based on the structural risk minimization principal, SVM defines an optimal hyperplane between samples of different class labels. The position of the hyperplane is adjusted so that the distance from the hyperplane to a nearest sample, or margin, is maximized.

Moreover, if the SVM cannot define any hyperplane that separates examples in linear space, it can use kernel functions to send examples to any kernel space where the hyperplane can separate examples. Although we can use any kernel function meeting Mercer's Theorem for SVM, we consider widely-used the linear and Gaussian radial basis function (RBF) kernels only in this research.

SVM-RFE

SVM is a powerful classification method but it has no feature selection method. Therefore, a wrapper-type feature selection method, SVM-RFE, was introduced [8]. SVM-RFE generates ranking of features by computing information gain during iterative backward feature elimination. The idea of information gain computation is based on Optimal Brain Damage (OBD) [12]. In every iterative step, SVM-RFE sorts the features in working set in the order of difference of the obejective functions and removes a feature with the minimum difference. Defining IG(k) as information gain when k-th feature is removed, overall iterative algorithm of SVM-RFE is shown in Algorithm 1.

ENSEMBLE and JOIN

SVM-RFE [8] has two parameters that need to be determined. The first parameter decides how many features should be used to obtain best performance. The second parameter specifies what portion of features should be eliminated in each iteration. To resolve this issue, a simple approach can be easily

Algorithm 1 SVM-RFE

Require: Feature lists R = [] and S = [1,..., P]

1: while S * [] do

2: Train a SVM with features in S

3: for all fc-th feature in S do

4: Compute IG(fc)

5: end for

6: e = arg minfc(IG(k))

7: R = [e, R]

8: S = S - [e]

9: end while

10: return R

implemented. First, we separate given training set into a partial training set and a hold-out set. Then, we apply Algorithm 2 with some parameter 'threshold'.

Score of each feature subset Ro is computed as

score(Ro) = err(Ro)+ || Ro ||/P

where err(Ro) is the error of SVM trained using Ro and tested with hold-out set. Using this method, we can obtain a feature subset R which yields reasonably small amount of error on trained dataset. Utilizing this algorithm as base, Jong et al. [10] proposed two methods, ENSEMBLE and JOIN to combine multiple rankings generated by SVM-RFE as in Algorithm 3 and 4.

In this paper, we used 25% of training set as hold-out set and used same sets of thresholds and cutoffs as in [10], i.e., {0.2, 0.3, 0.4, 0.5, 0.6, 0.7} and {1, 2, 3, 4, 5}.

Algorithm 2 SVM-RFE(threshold)

Require: Ranked feature lists R = [], Ri = [] where i = 1,..., P and S' = [1,..., P]

1: i = 1

2: while S' * [] do

3: Train an SVMs using a partial trainset with features in S'

4: for all features in S' do

5: Compute ranking of features as in SVM-RFE

6: end for

7: Rt = S'

8: Eliminate threshold percent of lesser-important features from S'

9: i = i + 1

10: end while

11: R = Ro where Ro yields minimum score on hold-out set.

12: return R

Algorithm 3 ENSEMBLE^, v2,.., vfc) 1: for threshold v e {v1, v2,..., vfc} do 2: Rv = SVM-RFE(v) 3: end for

4: return a majority vote classifier using SVMs trained by

Algorithm 4 JOIN(cutoff, v1, v2,..., vfc) 1: for threshold v e {v1, v2,..., vfc} do 2: Rv = SVM-RFE(v) 3: end for

4: R = features selected at least cutoff times in

{ Rv1,---, Rvfc }

5: return a SVM trained with R

Multiple SVM-RFE with bootstrap

Multiple SVM-RFE (MSVM-RFe) [9] is a recently introduced SVM-RFE-based feature selection algorithm. It exploits an ensemble of SVM classifiers and cross validation schemes to rank features. First, we make T subsam-ples from the original training set. Then, supposing that we have T SVMs trained using different subsamples, we calculate the corresponding discriminant information gain associated with each feature of each SVM. To compute this information gain, we use the same method as in SVM-RFE [8]. Exploiting the objective function of SVM, and its Lagrangian solution X, we can derive a cost function

J = (1 / 2)xTHx = xT1

where H is a matrix with elements yqyrK(xq, xr) and 1 is a N dimensional vector of ones while K( ) is a kernel func-

tion and 1 < q, r < N. Since we are looking for the subset of features that has the best discriminating power between classes, we compute the difference in cost function for each elimination of i-th input feature, leaving Lagrangian multipliers unchanged. Therefore, the ranking for the i-th feature of j-th SVM can be defined as

DJi = (1 / 2)AjTHAj - (1 / 2)AjTH(-i)Aj.

where H(-i) denotes that i-th feature was removed from all elements in H. Then, considering DJj as a weight vector of features for j-th SVM, we normalize all T weight vectors such as DJj = DJj/||DJj||. This gives us T weight vectors each with P elements. Here, each element in the vector stands for a information gain achieved by eliminating the corresponding feature. After normalizing weight vectors for each SVM, we can compute each feature's ranking score

ci = Vi/ai with v and ui defined as:

V = (1/ T )£ DJjt j=1

= jx (DJji - )2/(T -1)

The algorithm then applies this method to the training set with k-fold cross validation scheme. If we perform 5-fold cross validation and generate 20 subsamples in each fold, we will eventually have T = 100 SVMs to combine. The overall MSVM-RFE algorithm is described in Algorithm 5.

Algorithm 5 MSVM-RFE

Require: Ranked feature lists R = [] and S' = [1,..., P] 1: while S' * [] do

2: Train T SVMs using T subsamples with features in S' 3: for all j-th SVM 1 < j < T do 4: for all i-th feature 1 < i < P do 5: Compute DJ^ 6: end for

7: Compute DJj = DJj/||DJj|| 8: end for

9: for all feature l e S' do

10: Compute cl using Equation (1)

11: end for

12: e = arg minl(c(l)) where l e S'

13: R = [e, R]

14: S' = S' - [e]

15: end while

16: return R

One should note that original MSVM-RFE proposed in [9] uses cross-validation scheme when generating subsam-ples. However, we omitted this step because combining boosting into the original MSVM-RFE algorithm with cross-validation scheme is very complex and may confuse the purpose of this study.

Multiple SVM-RFE with boosting

When making subsamples, original MSVM-RFE uses the bootstrap approach [13]. This ensemble approach builds replicates of the original data set S by random re-sampling from S, but with replacement N times, where N is the number of examples. Therefore, each example (xn, yn) may appear more than once or not at all in a particular replicate subsample. Statistically, it is desirable to make every replicate differ as much as possible to gain higher improvement of the ensemble. The concept is both intuitively reasonable and theoretically correct. However, as the architecture of MSVM-RFE uses simple bootstrapping, it naturally follows that utilizing another popular ensemble method, boosting [14], instead of bootstrapping for two reasons. First, boosting outperforms bootstrapping on average [15,16], and secondly, boosting of SVMs generally yields better classification accuracy than bootstrap counterpart [17]. Therefore, to make use of ensemble of SVMs effectively, it may be worthwhile to use boosting instead of bootstrapping. For this reason, we applied Ada-Boost [14], a classic boosting algorithm, to MSVM-RFE algorithm instead of bootstrapping in this work.

Unlike the simple bootstrap approach, AdaBoost maintains weights of each example in S. Initially, we assign same value of weight to n-th example D1(n) = 1/N where 1 < n < N. Each iterative process consists of four steps. At first, the algorithm generates a bootstrap subsample according to weight distribution at t-th iteration Dt. Next, it trains an SVM using the subsample. Third, it calculate the error using the original example set S. Finally it updates the weight value so that the probability of cor-

rectly classified examples is decreased while that of incorrect ones is increased. This update procedure makes next bootstrap pick more incorrectly classified examples, i.e. difficult-to-classify examples than easy-to-classify ones. The iterative re-sampling procedure

MAKE_SUBSAMPLES() using AdaBoost algorithm is described in Algorithm 6.

Algorithm 6 MAKE_SUBSAMPLE

Require: S = {(xn, yn)}, D1(n) = 1/N, n = 1,..., N;

1: for j = 1 to T do

2: Build a bootstrap Bj = {(xn, yn)|n = 1,..., N} based on weight distribution Dj

3: Train a SVM hypothesis hj using Bj

4: t =£N=X Dj(n)[yn * hj(xn)] 5: if > 0.5 then 6: Goto line 2 7: end if

8: a = (1/2)ln((1 - j)/j), a e R

9: Dj+1(n) = (Dj(n)/Zj) x exp(-ajynhj(xn)) where Zj is a normalization factor chosen so that Dj+1 also be a probability distribution

10: end for

11: return Bj, a where 1 < j < T

In addition to modifying re-sampling method, we made a change in ranking criterion of original MSVM-RFE. In this MSVM-RFE with Boosting method, the weight vector DJj ofj-th SVM undergoes one more process between normalization and feature ranking score calculation. Since the contribution of each SVM in ensemble to the overall classification accuracy is unique, we multiply another weight factor to the normalized feature weight vector DJj. The new weight factor is obtained from the weight of hypothesis classifier calculated during the re-sampling process of AdaBoost. By multiplying this weight aj to DJj, we can grade the overall feature weight more coherently. The overall iterative algorithm of MSVM-RFE with AdaBoost is described in Algorithm 7.

Algorithm 7 MSVM-RFE with AdaBoost

Require: Ranked feature lists R = [] and S'= [1,..., P] 1: MAKE_SUBSAMPLES(Bt, a); t = 1,..., T 2: while S' * [] do

3: Train T SVMs using Bt, with features in set S'

4: Compute and normalize T weight vectors DJj as in MSVM-RFE where 1 < j < T

5: for j = 1 to T do

6: DJj = DJj x ln(aj)

7: end for

8: for all feature l e S' do

9: Compute the ranking score cl using Eq. (1)

10: end for

11: e = argminl(cl) where l e S'

12: R = [e, R]

13: S' = S' - [e]

14: end while

15: return R

Note that we took logarithm of hypothesis weights instead of raw values in order to avoid radical changes in ranking criterion. Since boosting algorithm overfits by nature and SVM, the base classifier, is relatively strong classifier, the error rate of hypothesis increases drastically as iteration in MAKE_SUBSAMPLES() progresses. We have

Table 1: Dataset Information

institution mass calcification

benign malignant benign malignant

MGH 482 365 381 323

WU 154 115 41 98

WFUSM 163 255 188 159

SHH 324 380 207 140

total 1 123 1 115 817 720

MGH = Massachussetts General Hospital; WU = Washington University at Saint Louis; WFUSM = Wake Forest University School of Medicine; SHH = Sacred Heart Hospital

Table 2: BI-RADS mammographie features feature type description or numeric value

mass shape no mass(0), round(1), oval(2), lobulated(3), irregular(4)

mass margin no mass(0), well circumscribed(i), microlobulated(2), obscured(3), ill-defined(4), spiculated(5)

calcification type no calc.(0), milk of calcium-like(i), eggshell(2), skin(3), vascular(4), spherical(5), suture(6), coarse(7), large rod-

like(8), round(9), dystrophic(iO), punctate(ii), indistinct(i 2), pleomorphic(l3), fine branching(i4)

calcification distribution no calc.(0), diffuse(i), regional(2), segmental(3), linear(4), clustered(5)

density i, 2, 3, 4

assessment i, 2, 3, 4, 5

density: 1 = sparser, 4 = denser;

witnessed this overfitting problem by preliminary experiment and solved the problem by taking logarithm to the hypothesis weight. Computation time of MSVM-RFE with boosting can also be explained here. From our experiments, we found that there is no significant difference between the original MSVM-RFE and MSVM-RFE with boosting as the number of subsamples generated by MAKE_SUBSAMPLES() decreases.

Lastly, unlike the conventional boosting algorithm application, we only exploit bootstrap subsamples generated by the algorithm and dismiss trained SVMs for the following reasons:

• We are primarily interested in feature ranking and not the aggregation of weak hypotheses.

• Since we are using SVM-RFE for eventual classification method, this require a certain criterion to pick appropriate number of features from different boosted models.

In preliminary experiments using same number of features and simple majority-voting aggregation, SVM-RFE using boosted models did not show significance in accuracy improvement. However, we could find some evidences that ensemble of SVMs can be useful in mammogram classification.

Results

In this section, we first describe dataset, features and experimental framework we used. Then we draw results of the experiments including analysis on them.

Dataset

The DDSM database provides about 2500 mammogram cases that were gathered from 1988 to 1999. Four U.S. medical institutions offered the data to construct DDSM. This includes Massachusetts General Hospital (MGH), Wake Forest University School of Medicine (WFUSM), Sacred Heart Hospital (SHH) and Washington University in St. Louis (WU). All mammogram cases we used in this paper contain one or more abnormalities which can be classified into benign or malignant group following their biopsy results. Table 1 summarizes the statistics of abnormalities from each digitizer type and institution.

Mammogram data from DDSM were gathered and pre-processed through the following steps. First, we extracted meta information from text file in the database. These features are based on Breast Imaging Reporting and Data System (BI-RADS) introduced by the American College of Radiology [18]. Table 2 summarizes these encoded features. We employed a rank ordering system proposed by other group when encoding these features [19]. Next, we computed statistical features that are popular in image processing community. The statistical features are computed using intensity level of pixels in the region of interest in each case. We used same features which are used in

Table 3: Comparison of kernels in terms of maximum Az value of mass dataset

kernel type MGH WU WUFSM SHH

8 22 8 22 8 22 8 22

linear

0.9039i 0.96664

0.90364

0.88597 5

0.9457i 0.95955

i0 0.5

0.92i59 0.92540

i0 0.075

0.857i 8 0.91906

i0 0.i 5

0.87i59 0.91671

i0 0.i

0.97 i50 0.97404

i0 0.5

0.97036

0.957i6 i0 0.05

Same tradeoff parameter value C is used for both linear and RBF kernels.

Table 4: Comparison of kernels in terms of maximum Az value of calcification dataset

kernel type MGH WU WUFSM SHH

В 22 В 22 В 22 В 22

linear 0.726S6 0.7262S 0.S99S I 0.90В70 0.74046 0.77s09 0.S9603 0.9270s

RBF 0.9i042 0.76В26 0.99i 92 0.SSISS 0.9362S 0.В9079 0.962В0 0.94В26

C I I0 I S I0 20 I0 I0

Y l.S 0.I I 0.0s 0.4 0.0s 0.ls 0.0s

Same tradeoff parameter value C is used for both linear and RBF kernels.

another study [6] and the exact formulas are described in [20]. We also normalized these statistical features after extracting because their raw values were too big compared to BI-RADS features and to facilitate SVM to train efficiently with respect to time.

Performance comparison

In sum, we prepared a total of 16 datasets each with 8 and 22 features, from each mass and calcification lesion of each institution. All SVM-RFE based methods are tested using 5-fold cross validation on each dataset. We computed area under Receiver Operating Characteristic (ROC) curves (Az) using the output of SVMs and feature ranking produced by each method.

Before comparing the methods explained in the previous section, we did some preliminary experiments comparing different kernels and parameters to find optimal kernel and parameters. The result of this experiment is summarized in Table 3 and Table 4. We used the best-performing

parameter and kernel (radial basis function, or RBF) from this experiment of this study.

The overall performance comparison result is summarized from Table 5 through Table 8. Note that numbers in parenthesis of JOIN methods are cutoff values used. Analyzing the result, it is clear that the MSVM-RFE based methods outperforms baseline classifiers, SVM and other SVM-RFE feature selection methods, ENSEMBLE and JOIN in the majority of cases although SVM-RFE dominated in 4 out of 16 datasets. Comparing the two MSVM-RFE based algorithms, we could find that MSVM-RFE with boosting can achieve better or at least competitive performance especially in datasets with 22 features. In 3 out of 4 mass datasets, MSVM-RFE with boosting outperformed any other methods under consideration. Although the original MSVM-RFE method yielded the best performance in 3 out of 4 calcification datasets, we think the MSVM-RFE with boosting has yet more margin to be

Table 5: Comparison of methods by maximum Az value using 8 features (Mass)

SVM-RFE

0.9SS2I 0.9621В

0.97247 0.97734

0.922S2 0.922S2

0.97401 0.97401

ENSEMBLE JOIN (1) JOIN (2) JOIN (3)

Join (4)

JOIN (5)

MSVM-RFE (bootstrap)

S I0 IS

0.72I02 0.77944 0.72I02 0.72I02 0.72I02 0.72I02

0.9SS2I 0.9SS2I 0.9S947 0.9S947

0.74SS9 0.SSIS7 0.7736S 0.7S4S4 0.7S4S4 0.7I I36

0.97247 0.97В51

0.974S7 0.9770S

0.67307 0.796SS 0.79200 0.79200 0.7S76S 0.67307

0.92423 0.922SS 0.922SS 0.923IS

0.94292 0.926S0 0.90262 0.S6SS7 0.S6S6I 0.7374S

0.97401 0.97S2S

0.97401 0.97401

MSVM-RFE (boost)

S I0 IS 20

0.9SS2I 0.9SS2I 0.9S947 0.9S947

0.97247 0.976I6 0.97247 0.973S7

0.923I4 0.92426

0.923I4 0.923I4

0.97401 0.97401 0.97401 0.97401

Numbers in parenthesis stands for cutoff value for JOIN method.

Table 6: Comparison of methods by maximum Az value using 8 features (Calcification)

SVM-RFE

0.9 ll82 0.9II96

0.98765

I.00000

0.94690 0.95100

0.96595 0.96595

ENSEMBLE JOIN (1) JOIN (2)

Join (3)

JOIN (4)

Join (5)

MSVM-RFE (bootstrap)

5 l0 l5 20

0.539l 5 0.67508 0.57971 0.57971 0.54571 0.54571

0.9 ll82 0.9 ll82 0.9 ll82 0.9 ll82

0.695l2 0.7l 655 0.7294l 0.7294l 0.695l2 0.695l2

0.98765 0.98765 0.98765 0.98765

0.56583 0.83947 0.76157 0.62686 0.62686 0.54077

0.95326

0.95168 0.94690 0.94690

0.9l392 0.93422 0.88733 0.73542 0.72464 0.66210

0.97S6S

0.96595 0.96757 0.97348

MSVM-RFE (boost)

5 l0 l5 20

0.9 ll82 0.9 ll82 0.9 ll82 0.9 ll82

0.98765 0.99259 0.99429 0.98765

0.94690 0.94690 0.94690 0.94690

0.96595 0.96595 0.96595 0.96595

Numbers in parenthesis stands for cutoff value for JOIN method.

improved as we already mentioned in the previous chapter. Any method that can effectively exploit the trained SVMs during feature selection progress may be the future key improvement for MSVM-RFE with boosting.

Conclusion

In this paper, a new SVM-RFE based feature selection method was proposed. We conducted experiments on real

world clinical data, and compared our method with baseline and other feature selection methods using SVM-RFE. Results show that our method outperforms in some cases and is at least competitive to others in other cases. Therefore, it can be a possible alternative to SVM-RFE or the original MSVM-RFE. Future works include investigation of specific methods to effectively combine models trained

Table 7: Comparison of methods by maximum Az value using 22 features (Mass)

T MGH WU WFUSM SHH

SVM-RFE

0.88805 0.88849

0.93642 0.94173

0.92474 0.93037

0.94998 0.94998

ENSEMBLE JOIN (1) JOIN (2) JOIN (3) JOIN (4) JOIN (5)

MSVM-RFE (bootstrap)

5 l0 l5 20

0.81490 0.86728 0.83034 0.75098 0.74270 0.68776

0.89720 0.88833

0.S9920

0.890l 4

0.90299 0.92278 0.93886 0.873l2 0.74262 0.7l3l6

0.93729 0.93666 0.93746 0.94290

0.803l7 0.87638 0.89597 0.82694 0.66948 0.66948

0.92664 0.92972 0.93000 0.92986

0.86l 55 0.90789 0.85l 32 0.83834 0.83834 0.80802

0.95087 0.95016 0.95076 0.95066

MSVM-RFE (boost)

5 l0 l5 20

0.88993 0.88805 0.89092 0.88805

0.93987 0.943I5

0.94204 0.94197

0.935SI

0.928l2 0.92789 0.92758

0.94998 0.94998 0.94998

0.95245

Numbers in parenthesis stands for cutoff value for JOIN method.

Table 8: Comparison of methods by maximum Az value using 22 features (Calcification)

SVM-RFE

0.77497 0.77497

0.91710 0.93436

0.S973S 0.S9Ss9

0.9494s 0.9s332

ENSEMBLE JOIN (1) JOIN (2) JOIN (3) JOIN (4) JOIN (5)

MSVM-RFE (bootstrap)

s 10 Is 20

0.6S9s I 0.7s2s9 0.72296 0.70SI s 0.sS6s6 0.s3s20

0.77497

0.77В26

0.77497 0.77497

0.76647 0.92326 0.S2307 0.76647 0.69779 0.63SsS

0.91710 0.91710 0.92193 0.9330s

0.726s0 0.SI433 0.729S7 0.700s9 0.6s667 0.6s667

0.S99SS 0.S97S6 0.S973S

0.90S07

0.Ss677 0.9I3s2 0.S0400 0.67s9S 0.ss964 0.si203

0.9S379

0.9s330 0.9s2s0 0.9s267

MSVM-RFE (boost)

s 10 Is 20

0.77727 0.77497 0.77497 0.77497

0.92097 0.93063 0.923s2 0.9210s

0.S9S4S 0.90I0S 0.90133 0.S99s7

0.9494s 0.9s292 0.9si 36 0.9s2s6

Numbers in parenthesis stands for cutoff value for JOIN method.

during the feature selection process and ways to combine feature subsets generated from individual SVM-RFE instances.

Competing interests

The authors declare that they have no competing interests. Authors' contributions

SY carried out the study, designed and implemented the algorithms, conducted experiments and drafted this manuscript. SK supervised and instructed all research progress, and participated in the algorithm design and critical analysis of results. Both authors read and approved the final manuscript.

Acknowledgements

The work of SK was supported by the Special Research Grant of Sogang University 200811028.01.

This article has been published as part of BMC Medical Informatics and Decision Making Volume 9, Supplement 1, 2009: 2008 International Workshop on Biomedical and Health Informatics. The full contents of the supplement are available online at http://www.biomedcentral.com/l472-6947/ 9?issue=Sl.

References

1. American Cancer Society: Cancer Facts and Figures. American Cancer Society, 250 Williams Street, NW, Atlanta, GA; 2008.

2. Elmore J, Armstrong K, Lehman C, Fletcher S: Screening for breast cancer. The Journal of the American Medical Association 2005, 293:1245-1256.

3. Lo J, Bilska-Wolak A, Baker J, Tourassi G, Floyd C, Markey M: Computer-Aided Diagnosis in breast imaging: Where do we go after detection? In Recent Advances in Breast Imaging, Mammography

and Computer-Aided Diagnosis of Breast Cancer Edited by: Suri J, Ran-gayyan R. SPIE Press; 2006:87l-900.

4. Fischer E, Lo J, Markey M: Bayesian networks of BI-RADS descriptors for breast lesion classification. Proc of the 26th IEEE EMBS, San Francisco, CA, USA 2004, 2:303l-3034.

5. Wei L, Yang Y, Nishikawa R, Jiang Y: A Study on Several Machine-Learning Methods for Classification of Malignant and Benign Clustered Microcalcifications. IEEE Transactions on Medical Imaging 2005, 24:37l-380.

6. Panchal R, Verma B: Characterization of Breast Abnormality Patterns in Digital Mammograms Using Auto-associator Neural Network. In ICONIP (3), Volume 4234 of Lecture Notes in Computer Science Edited by: King I, Wang J, Chan L, Wang DL. Springen 2006:l 27-l36.

7. Land WH Jr, Mckee D, Velazquez R, Wong L, Lo J, Anderson F: Application of Support Vector Machines to breast cancer screening using mammogram and clinical history data. Proc SPIE, Volume 5032 of Medical Imaging 2003: Image Processing 2003:546-556.

8. Guyon I, Weston J, Barnhill S, Vapnik V: Gene Selection for Cancer Classification using Support Vector Machines. Machine Learning 2002, 46(l-3):389-422.

9. Duan K, Rajapakse J, Wang H, Azuaje F: Multiple SVM-RFE for gene selection in cancer classification with expression data. IEEE Transactions on Nanobioscience 2005, 4(3):228-234.

10. Jong K, Marchiori E, Sebag M, Vaart A van der: Feature selection in proteomic pattern data with support vector machines. Proceedings of the 2004 IEEE Symposium on Computational Intelligence in Bioinformatics and Computational Biology (CIBCB) 2004:41 -48.

l l. Heath M, Bowyer K, Kopans D, Moore R, Kegelmeyer W: The Digital Database for Screening Mammography. In Proc of the 5th IWDM Edited by: Yaffe M. Medical Physics Publishing; 200l:2l2-2l8.

12. LeCun Y, Denker JS, Solla SA: Optimal Brain Damage. In Advances in Neural Information Processing Systems Morgan Kaufmann; 1990:598-605.

13. Efron B: Bootstrap Methods: Another Look at the Jackknife. The Annals of Statistics l979, 7: l-26.

14. Freund Y, Schapire RE: A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting. Journal of Computer and System Sciences l997, 55: ll9-l39.

15. Bauer E, Kohavi R: An Empirical Comparison of Voting Classification Algorithms: Bagging, Boosting, and Variants. Machine Learning l999, 36(1-2): 105-l 39.

16. Tan PN, Steinbach M, Kumar V: Introduction to Data Mining Addison Wesley; 2005.

17. Kim HC, Pang S, Je HM, Kim D, Bang S: Pattern Classification Using Support Vector Machine Ensemble. Pattern Recognition 2002, 2:1051-4651.

18. American College of Radiology: Breast Imaging Reporting and Data System (BI-RADS) Reston, VA, USA: American College of Radiology; 1998.

19. Lo J, Gavrielides M, Markey M, Jesneck J: Computer-aided classification of breast microcalcification clusters: Merging of features from image processing and radiologists. In Medical Imaging 2003: Image Processing Volume 5032. Edited by: Sonka M, Fitz-patrickJ. SPIE Press; 2003:882-889.

20. Zhang P, Verma B, Kumar K: Neural vs. statistical classifier in conjunction with genetic algorithm based feature selection. Pattern Recognition Letters 2005, 26(7):909-919.

Publish with BioMedCentral and every scientist can read your work free of charge

"BioMed Central will be the most significant development for disseminating the results of biomedical research in our lifetime." Sir Paul Nurse, Cancer Research UK

Your research papers will be:

• available free of charge to the entire biomedical community

• peer reviewed and published immediately upon acceptance

• cited in PubMed and archived on PubMed Central

• yours — you keep the copyright

Submit your manuscript here: f J BioMedcentral

http://www.biomedcentral.com/info/publishing_adv.asp ^