Breast Cancer Diagnosis Using Artificial Neural Networks With Extreme .
(IJARAI) International Journal of Advanced Research in Artificial Intelligence,Vol. 3, No. 7, 2014Breast Cancer Diagnosis using Artificial NeuralNetworks with Extreme Learning TechniquesChandra Prasetyo UtomoAan KardianaRika YuliwulandariYARSI E-Health Research CenterFaculty of Information TechnologyYARSI University,Jakarta, IndonesiaYARSI E-Health Research CenterFaculty of Information TechnologyYARSI University,Jakarta, IndonesiaYARSI Genetics Research CenterFaculty of MedicineYARSI University, Jakarta,IndonesiaAbstract—Breast cancer is the second cause of dead amongwomen. Early detection followed by appropriate cancertreatment can reduce the deadly risk. Medical professionals canmake mistakes while identifying a disease. The help of technologysuch as data mining and machine learning can substantiallyimprove the diagnosis accuracy. Artificial Neural Networks(ANN) has been widely used in intelligent breast cancerdiagnosis. However, the standard Gradient-Based BackPropagation Artificial Neural Networks (BP ANN) has somelimitations. There are parameters to be set in the beginning, longtime for training process, and possibility to be trapped in localminima. In this research, we implemented ANN with extremelearning techniques for diagnosing breast cancer based on BreastCancer Wisconsin Dataset. Results showed that ExtremeLearning Machine Neural Networks (ELM ANN) has bettergeneralization classifier model than BP ANN. The development ofthis technique is promising as intelligent component in medicaldecision support systems.Keywords—breast cancer; artificial neural networks; extremelearning machine; medical decision support systemsI.INTRODUCTIONThe out of control development of cells in an organ iscalled tumors that can be cancerous. There are two kinds oftumors, benign and malignant. Benign or non-cancerous tumorsare not spreading and are not life intimidating. In the otherhand, malignant or cancerous tumor are expanding and lifethreatening [1]. Malignant breast cancer is defined when thegrowing cells are in the breast tissue. Breast cancer is thesecond overall cause of mortality among women and the firstcause of dead among them between 40 and 55 ages [2].Regular breast cancer diagnosis followed by appropriatecancer treatment can reduce the unwilling risk. It is suggestedto do tumor evaluation test every 4-6 weeks. Based on thatreason, benign and malignant detection based on classificationfeatures become very important [3].Careful diagnosis in early detection has been proven tolessen the dead rate because of breast cancer [4]. Depend on theexpertise, mistakes can be made by medical professionals whileidentifying a disease. With the help of technology such as datamining and machine learning, diagnosis can be more accurate(91.1%) when related to a diagnosis made by an experienceddoctor (79.9%) [5].ANN is one of the best artificial intelligence techniques forcommon data mining tasks, such classification and regressionproblems. A lot of research showed that ANN delivered goodaccuracy in breast cancer diagnosis. However, this method hasseveral limitations. First, ANN has some parameters to betuned in the beginning of training process such as number ofhidden layer and hidden nodes, learning rates, and activationfunction. Second, it takes long time for training process due tocomplex architecture and parameters update process in eachiteration that need expensive computational cost. Third, it canbe trapped to local minima so that the optimal performancecannot be guaranteed. Numerous efforts had been attempted toget the solutions of neural networks limitations. Huang andBabri [6] proved that Single Hidden Layer Neural Networks(SFLN) with tree steps extreme learning process that calledELM can solve that problems.In this paper, we revealed the implementation of artificialneural networks with extreme learning techniques in breastcancer diagnosis. The dataset used for experiments was BreastCancer Wisconsin Dataset that was obtained from theUniversity of Wisconsin Hospital, Madison from Dr. WilliamH. Wolberg [7]. We compared the perfomance of ELM withconventional BP ANN with gradient descent based learningalgorithms. Sensitivity, specificity, and accuracy were used asperformance measurements. Results showed that ELM ANNgenerally produced better result than BP ANN.The rest of this paper is organized as the following. Section2 is dedicated as literature review. In this section, brief reviewof previous works in breast cancer diagnosis are presented. InSection 3, the concept, mathematical model, and trainingprocess of extreme learning machine are explained. In Section4, experiments, results, and analysis are provided. Finally,conclusions and future works are given in Section 5.II. LITREATURE REVIEWThe uses of classification systems in medical diagnosis,including breast cancer diagnosis, are growing rapidly.Evaluation and decision making process from expert medicaldiagnosis is key important factor. However, intelligentclassification algorithm may help doctor especially inminimizing error from unexperienced practitioners [3].Several techniques have been deployed to predict andrecognize meaningful pattern for breast cancer diagnosis. RyuaThis research is funded by YARSI University in the Internal ResearchGrant Schema 2013/2014.10 P a g ewww.ijarai.thesai.org
(IJARAI) International Journal of Advanced Research in Artificial Intelligence,Vol. 3, No. 7, 2014[8] developed data classification method, called isotonicseparation. The performances were compared against supportvector machines, learning vector quantization, decision treeinduction, and other methods based on two-breast cancer dataset, sufficient and insufficient data. The experiment resultsdemonstrated that isotonic separation was a practical tool forclassification in the medical domain.Hybrid machine learning method was applied by Sahan [9]in diagnosing breast cancer. The method hybridized a fuzzyartificial immune system with k-nearest neighbour algorithm.The hybrid method delivered good accuray in WisconsinBreast Cancer Dataset (WBCD). They believe it can also betested in other breast cancer diagnosis problems.Comprehensive view of automated diagnostic systemsimplementation for breast cancer detection was provided byUbeyli [10]. It compared the performances of multilayerperceptron neural network (MLPNN), combined neuralnetwork (CNN), probabilistic neural network (PNN), recurrentneural network (RNN) and support vector machine (SVM). Theaim of that works was to be a guide for a reader who wants todevelop this kind of systems.Numerous combinations and hybrid systems used neuralnetworks as a component. However, since almost all of theemployed neural networks are conventional gradient descentBP ANN, the novel or hybrid method still suffered the neuralnetworks drawbacks that were mentioned in the previoussection.III. EXTREME LEARNING MACHINEHuge efforts had been attempted to solve the weaknesses ofBP ANN. Huang and Babri [6] demonstrated that single hiddenlayer feedforward neural networks (SLFN) with at most mhidden nodes was capable to estimate function for m differentvectors in training dataset.Given m instances in D {(x(k), t(k)) x(k) ϵ Rn, t(k) ϵ Rp, k 1,.,m} as training dataset where x(k) [x1(k), x2(k), ., xn(k)]T asfeatures and t(k) [t1(k), t2(k), ., tp(k)]T as target. A SLFN with Mnumber of hidden nodes, activation function g(x) in hiddennodes, and linear activation function in output nodes ismathematically wrote as:βi [βi1, βi2, . . . , βip]T,(3)wi x(k) is the inner product of wi and x(k),bi is the bias of the i-th hidden node,o(k) ϵ Rp is the output of neural network for k-th vector.SLFN can approximate m vectors means that there are existwi, βi, and bi, such that:(4)(5)Equation (5) can be written as:,(6)whereH є Rm x M is the hidden layer output matrix of the neuralnetworks. g ( w1 x (1) b1 ) g ( wM x (1) bM ) g ( w1 x ( m ) b1 ) g ( wM x ( m ) bM ) H β є RMxp(7)is the weights between hidden and output layers 1T β , T M (8)T є Rm x p is the target values of m vectors in training dataset t (1) T T , t ( m ) T (1)wherewi ϵ Rn is the weights between the input nodes and the i-thhidden nodewi [wi1, wi1, . . . , win ]T,(2)βi ϵ Rp is the weights between the i-th hidden node and theoutput nodes(9)In the traditional gradient descent based learning algorithm,weights wi which was connecting the input layer and hiddenlayer and biases bi in the hidden nodes were needed to beinitialized and tuned in every iteration. This was the mainfactor which often made training process of neural became timeconsuming and the trained model may not reach globalminima.11 P a g ewww.ijarai.thesai.org
(IJARAI) International Journal of Advanced Research in Artificial Intelligence,Vol. 3, No. 7, 2014Huang [11] proposed minimum norm least-squares solutionof SLFN which didn’t need to tune those parameters. TrainingSLFN with fixed input weights wi and the hidden layer biases biwas similar to find a least square solutionof the linearsystem:Cancer Wisconsin Dataset obtained from the University ofWisconsin Hospital, Madison from Dr. William H. Wolberg[7]. The data has 699 instances with 10 attributes plus the classattributes. The class distribution are 65.5% (458 instances) forbenign and 34.5% (241 instances) for malignant. The attributeinformation can be seen in TABLE I.–TABLE I.(10)The smallest norm least squares solution of that linearsystem was(11)wherewas the Moore-Penrose generalized inverse ofmatrix H. This solution had three important properties whichwere minimum training error, smallest norm of weights, andunique solution which is.The above minimum norm least-square solution for SLFNwas called extreme learning machine (ELM). Given minstances in training dataset D {(x(k), t(k)) x(k) є Rn, t(k) є Rp, k 1,.,m}, activation function g(x), and number of hidden nodeM. The training process of ELM is the the following:1.2.3.Randomly set input-hidden layer weights wi and biasbi, i 1, ,M.Compute the matrix of hidden layer output HCompute the hidden-output layer weightswhere T [t(1), , t(m)].forBased on that definition, there are three main differencesbetween BP ANN and ELM ANN. First, BP ANN needs totuning several parameters, such as number of hidden nodes,learning rates, momentum, and termination criteria. On theother hand, ELM ANN is a simple tuning free algorithm. Theonly one to be defined is number of hidden nodes. Second, BPANN works only for differentiable activation functions inhidden and output nodes while ELM ANN can use bothdifferentiable and undifferentiable activation functions.#1234567891011A. ExperimentsThe experiments consisted of three main steps, which weredata gathering, data preprocessing, and performanceevaluating. The dataset used in this experiment was BreastAttributesSample Code NumberClump ThicknessUniformity of Cell SizeUniformity of Cell ShapeMarginal AdhesionSingle Epithelial Cell SizeBare NucleiBland ChromatinNormal NucleoliMitosesClassDomainid number1 – 101 – 101 – 101 – 101 – 101 – 101 – 101 – 101 – 102: benign; 4: malignantIn the second step, the raw dataset was preprocessed toproduce well-from data that suitable for training and testingprocess. The first attribute, sample code number, was removedbecause it was not relevant to the diagnosis. The next nineattributes were normalized into [-1, 1] and used as predictor.The last attribute was transformed to 0 (benign) and 1(malignant) such that it can be properly fitted to the standardBP ANN and ELM ANN implementation.The method in this experiment was k-fold crossvalidationwith k 5. This means, the data were randomly divided into 5partitions. There were 5 experiments. In the each experiment, apartition was used as testing data and the rest partitions weretreated as training data.The standard performance measurement for classificationproblem was accuracy. However, since the class distributionwas not balanced, it was important to use specificity andsensitivity as supplementary measurements. In addition, tominimize the effect of random generated weights in BP ANNand ELM ANN, each experiment was run three times and theaverage results were noted.Finally, BP ANN get trained model which has minimumtraining error so that there is a possibility to finish in localminima. On the other hand, ELM ANN get trained modelwhich has minimum training error and smallest norm of weightso that it can produce better generalization model and reachglobal minima [12].IV. EXPERIMENTS, RESULTS, AND ANALYSISThis section discussed about experimental design,generated results, and analytical process in order to get validconclusion.ATTRIBUTE INFORMATION(13)(14)(15)B. Results and AnalysisWith 5-fold crossvalidation method and each experimentwere run three times, there will be 15 experiments in total. Thewhole steps had been done in computer with Intel Core i3,4096MB RAM, and Windows 7 OS. The results of ELM aregiven in TABEL II.12 P a g ewww.ijarai.thesai.org
(IJARAI) International Journal of Advanced Research in Artificial Intelligence,Vol. 3, No. 7, 850,9640,9710,956In order to compare the ELM ANN performances, trainingand testing with BP ANN were conducted with identicalexperimental design. The performances of BP ANN can beseen in TABLE III.TABLE III.Experiments#1#2#3#4#5Running123123123123123BP ANN 050,9490,9490,942To have clear view between the performances of ELMANN and BP ANN, the results were transformed to graphicalcharts. In each performance measurement, the average valueswere computed in each experiment. Fig 1 shows thecomparison of average sensitivity rates between BP ANN andELM ANN. The comparison of specificity rates were given inFig 2 while accuracy rates can be seen in Fig 3.Average Sensitivity Rates#1Running1231231231231231.0000.8000.600BP ANN0.400ELM ANN0.2000.00012345Experiments #Fig. 1. Average sensitivity rates of BP ANN dan ELMAverage Specificity RatesExperimentsELM PERFORMANCES1.0000.8000.600BP ANN0.400ELM ANN0.2000.00012345Experiments #Fig. 2. Average specificity rates of BP ANN dan ELMAverage Accuracy RatesTABLE II.1.0000.8000.6000.400BP ANN0.200ELM ANN0.00012345Experiments #Fig. 3. Average accuracy rates of BP ANN dan ELMBased on Fig 1 and Fig 3, we can see that ELM ANN wassuperior compared to BP ANN. ELM ANN has betterperformances in term of sensitivity and accuracy in allexperiments.13 P a g ewww.ijarai.thesai.org
(IJARAI) International Journal of Advanced Research in Artificial Intelligence,Vol. 3, No. 7, 2014However, in term of specificity, BP ANN has betterperformance in three experiments which were experiment #1,#2, and #5.To get general conclusion, overall comparison need to becomputed. In each performance measurement, commulativeaverage rates were matched. Fig 4 displays the wholesensitivity, specificity, and accuracy average rates between BPANN and ELM ANN. Result showed that, generally ELMANN were better than BP ANN.0.9481.0000.9000.980 yBP ANNSpecificityAccuracyELM ANNFig. 4. Overall average rates between BP ANN dan ELMV. CONLUSION AND FUTURE WORKSThe performances of ELM ANN were generally better thanBP ANN in breast cancer diagnosis. Although the specificityrate was slightly lower than BP ANN, it can be clearly seenthat ELM ANN remarkably improved the sensitivity andaccuracy rates. Based on these results, we can conclude thatELM ANN has better generalization model than BP ANN indiagnosing breast cancer based on Breast Cancer WisconsinDataset.There are some necessary works to be done in near future.First, it is important to communicate the model with domainexpert. The hybrid of ELM ANN with Decision Tree or anyother technique that can produce meaningful knowledgerepresentation will be promising. Second, to make intelligentdiagnosis tool that can be used by end user, it is necessary todevelop interactive user interface.The development of interfaces in mobile, desktop, or webapplication may be useful. Third, there are new cases addedregularly in the hospital. Developing intelligent diagnosissystems that can not only learn from available data inrepositories but also from newly available data will be required.REFERENCEST. S. Subashini, V. Ramalingam, and S. Palanivel, “Breast massclassification based on cytological patterns using RBFNN and SVM,”Expert Syst. Appl., vol. 36, no. 3, pp. 5284–5290, Apr. 2009.[2] TheAmericanCancerSociety, “What is Breast Cancer.” .[3] M. F. Akay, “Support vector machines combined with feature selectionfor breast cancer diagnosis,” Expert Syst. Appl., vol. 36, no. 2, pp. 3240–3247, Mar. 2009.[4] D. West, P. Mangiameli, R. Rampal, and V. West, “Ensemble strategiesfor a medical diagnostic decision support system: A breast cancerdiagnosis application,” Eur. J. Oper. Res., vol. 162, no. 2, pp. 532–551,Apr. 2005.[5] R. W. Brause, “Medical Analysis and Diagnosis by Neural Networks,”in Proceeding ISMDA ’01 Proceedings of the Second InternationalSymposium on Medical Data Analysis, 2001, pp. 1–13.[6] G. B. Huang and H. A. Babri, “Upper bounds on the number of hiddenneurons in feedforward networks with arbitrary bounded nonlinearactivation functions.,” IEEE Trans. Neural Netw., vol. 9, no. 1, pp. 224–9, Jan. 1998.[7] W. H. Wolberg and O. L. Mangasarian, “Multisurface method of patternseparation for medical diagnosis applied to breast cytology,” inProceedings of the National Academy of Sciences, U.S.A., Volume 87,December 1990, pp. 9193–9196.[8] Y. U. Ryu, R. Chandrasekaran, and V. S. Jacob, “Breast cancerprediction using the isotonic separation technique,” Eur. J. Oper. Res.,vol. 181, no. 2, pp. 842–854, Sep. 2007.[9] S. Sahan, K. Polat, H. Kodaz, and S. Güneş, “A new hybrid methodbased on fuzzy-artificial immune system and k-nn algorithm for breastcancer diagnosis.,” Comput. Biol. Med., vol. 37, no. 3, pp. 415–23, Mar.2007.[10] E. D. Übeyli, “Implementing automated diagnostic systems for breastcancer detection,” Expert Syst. Appl., vol. 33, no. 4, pp. 1054–1062,Nov. 2007.[11] G.-B. Huang, Q.-Y. Zhu, and C.-K. Siew, “Extreme learning machine:Theory and applications,” Neurocomputing, vol. 70, no. 1–3, pp. 489–501, Dec. 2006.[12] C. P. Utomo, “The Hybrid of Classification Tree and Extreme LearningMachine for Permeability Prediction in Oil Reservoir,” Int. J. Comput.Sci. Issues, vol. 10, no. 1, pp. 52–60, 2013.[1]14 P a g ewww.ijarai.thesai.org
Malignant breast cancer is defined when the growing cells are in the breast tissue. Breast cancer is the second overall cause of mortality among women and the first cause of dead among them between 40 and 55 ages [2]. Regular breast cancer diagnosis followed by appropriate cancer treatment can reduce the unwilling risk.
4 Breast cancer Breast cancer: A summary of key information Introduction to breast cancer Breast cancer arises from cells in the breast that have grown abnormally and multiplied to form a lump or tumour. The earliest stage of breast cancer is non-invasive disease (Stage 0), which is contained within the ducts or lobules of the breast and has not spread into the healthy breast tissue .
Alcohol before breast cancer diagnosis and breast cancer mortality . 81 Alcohol less than 12 months after diagnosis and breast cancer mortality . 86 Alcohol intake 12 months or more after primary breast cancer diagnosis and breast . Before diagnosis BMI and cardiovascular disease mortality . 317 BMI less than 12 months after .
Breast cancer development In the United States, breast cancer is the most common cancer diagnosed in women (excluding skin cancer). Men may also develop breast cancer, but less than 1% of all people with breast cancer are men. Breast cancer begins when healthy cells in the breast change and grow uncontrollably, forming a mass called a tumor.
breast cancer, metastases, advanced breast cancer, secondary tumours, secondaries or stage 4 breast cancer. For most people with secondary breast cancer in the brain, breast cancer has already spread to another part of the body such as the bones, liver or lungs. However, for some people, the brain may be the only area of secondary breast cancer.
All breast cancer cases ER Positive ER Negative 0246810 B: Year after diagnosis 0 204 060 P e rc ent annual haz ard rate All breast cancer cases 02468 C: Year after diagnosis 040 80 % s u rv i v i ng free of breast c ancer death Inflammatory breast cancer 02468 D: Year after diagnosis 020 40 60 P e rc ent annual haz ard rate Inflammatory breast .
Beyond Breast Cancer Awareness: A panel discussion on advancements in breast cancer genetics, research and treatment Join our panel of experts who will discuss research on the molecular level of breast cancer, clinical trials, advanced breast cancer treatments, and hereditary and other risk factors for developing breast cancer. Hosted
5 yrs of anti-hormone therapy reduces the risk of: – breast cancer coming back somewhere else in the body (metastases / secondaries) – breast cancer returning in the same breast – a new breast cancer in the opposite breast – death from breast cancer The benefits of anti-hormone therapy last well
EMC Overview 43 Immunity Issues Can Exist Due To The Following Most of today’s electrical and electronic systems rely on active devices such as microprocessors and digital logic for: –Control of system functions. –User convenience / features. –Legislated system requirements (such as mobile telephone location reporting). With today’s vast networks for data communication there .
