Indonesian J our nal of Electrical Engineering and Computer Science V ol. 25, No. 1, January 2022, pp. 580 ∼ 588 ISSN: 2502-4752, DOI: 10.11591/ijeecs.v25.i1.pp580-588 ❒ 580 A utomated br east cancer detection system fr om br east mammogram using deep neural netw ork Suneetha Chittineni 1 , Sai Sandeep Edara 2 1 Department of Computer Applications, R. V . R. and J. C. Colle ge of Engineering, Cho wda v aram, Guntur , India 2 Department of Computer Science and Engineering, R. V . R. and J. C. Colle ge of Engineering, Cho wda v aram, Guntur , India Article Inf o Article history: Recei v ed Apr 8, 2021 Re vised No v 23, 2021 Accepted No v 26, 2021 K eyw ords: Breast cancer Deep neural netw ork Hybridization Mammograph y Thermograph y ABSTRA CT All o v er the w orld breast cancer is a major disease which mostly af fects the w omen and it may also cause death if it is not diagnosed in its early stage. But no w adays, se v eral screening methods lik e magnetic resonance imaging (MRI), ultrasound imag- ing, thermograph y and mammograph y are a v ailable to detect the breast cancer . In this article mammograph y images are used to detect the breast cancer . In mammogra- ph y image the cancerous lumps/microcalcications are seen to be tin y with lo w con- trast therefore it is dif cult for the doctors/radiologist to detect it. Hence, to help the doctors/radiologist a no v el system based on deep neural netw ork is introduced in this article that detects the cancerous lumps/microcalcica tions automatically from the mammogram images. The system acquires the mammographic images from the mammographic image analysis society (MIAS) data set. After pre-processing these images by 2D median image lter , c ancerous features are e xtracted from the images by the h ybridization of con v olutional neural netw ork with rat sw arm optimization al- gorithm. Finally , the breast cancer patients are classied by inte grating random forest with arithmetic optimization algorithm. This system identies the breast cancer pa- tients acc urately and its performance is relati v ely high compared to other approaches. This is an open access article under the CC BY -SA license . Corresponding A uthor: Suneetha Chittineni Department of Computer Applications, R. V . R. and J. C. Colle ge of Engineering Cho wda v aram, Guntur , India Email: suneethachittineni@gmail.com 1. INTR ODUCTION One of the most common diseases that af fect w omen in recent years is the breast cancer [1]. In the latest s urv e y tak en by w orld health or g anization (WHO) it is predicted that by 2025 in the w orld there are 19.3 million victims af fected by breast ca n c er . Breast cancer is a condition in which cells gro w out of control, result- ing in a tumour that can spread throughout the body . Although the specic causes of breast cancer are unkno wn, researchers belie v e that aberrant cell gro wth is caused by a combination of genes, lifestyle, en vironment, and hormones [2]. This breast cancer must be detected in its early stage otherwise it may cause death. Hence, there are numerous medical imaging techniques lik e m agnetic resonance imaging (MRI), ultrasound imaging, thermograph y and mammograph y are a v ailable to identify the breast cancer [3]. But, diagnosing breast cancer at its early stage becomes a challenging w ork to the medical e xperts lik e doctors and radiologists. In magnetic resonance imaging (MRI) the breast images are captured from 3D vie w . It emplo ys a non-ionizing radiation [4]. But the rate of MRI is high and it is dif cul t to dif ferentiate the normal lumps J ournal homepage: http://ijeecs.iaescor e .com Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 581 and cancerous lumps from the breast MRI. The breast ultrasound produces less accurate results for patients with dense breast [5]. The result of images is bas ed on the e xpert who is taking the ultrasound for the patient. So man y times it produces high f alse positi v e rate that leads to unnecessary biopsy [6]. Breast thermograph y utilizes the infrared cameras to capture the breast images [7]. The cam era has an inb uilt infrared sensor that helps to record the temperature of the breast. Based on the v ariation in the tem perature the breast cancer is detected. If an y cold pressure w as applied to the breast then the original breast temperature changes and produces f alse result. T o a v oid these limitations in this study mammographic images are used to identify the breast cancer in its early stage. Mammograph y is one of the commonly used and belie v ed methods to identify the lumps in the breast. The mammograph y images sho w the presence of cancerous lumps/ microcalcications in the brea st [8]. The cancerous lumps in the mammograph y images are tin y in size and its image contrast is lo w . So it is hard for the doctors/radiologists to detect the microcalcications/cancerous lumps in the mammographic im- ages. Hence, to e ase the w ork of the doctors/radiologists, a no v el system is proposed that detects the cancerous lumps in the breast from the mammographic images. The proposed dee p neural netw ork system acquires the mammograph y images from the mammographic image analysis society (MIAS) image datas et. The con v olu- tional neural netw ork (CNN) algorithm, inte grated with rat sw arm optimization e xtracts the features of breast cancer from the mammographic images. The features are e xtracted by tuning the parameters of CNN and there by updating the position of the rat sw arm optimization. Then the e xtracted features are classied using the classier random forest inte grated with the arithmetic optimization algorithm. The classier is designed by the arithmetic optimization algorithm which helps to a v oid the reasoning problem occur in the output of CNN. Figure 1 describes the block diagram of the proposed system. Figure 1. Block diagram of the proposed system This article is planned as follo ws. Literature re vie w based on mammograph y images, e xtracting the features and classifying the breast cancer is e xplained in section 2. Section 3 e xplains the background models used i n the proposed approach. Section 4 describes the proposed method and the algorithms used in the pro- posed method. In section 5, e xperimental results with simulation are e xplained and discussed. The conclusion of this article and its future w ork is discussed in section 6. 2. LITERA TURE REVIEW Cao et al. [9] introduces a no v el con v olutional neural netw ork (CNN) frame w ork to identify the breast cancer from the ultrasound images. This frame w ork contains se v eral object detection and classication approaches to identify the tumour . It rst detects the presence of tumour in the breast ultrasound image and then classies the type of the tumour by the CNN frame w ork. In this, undertting problem occurs while nding the malignant lumps and only less parameter are considered to identify the tumour . Singh and Singh [10] combine and impro v e se v eral e xisting approaches in se gmentation, feature selection, feature e xtraction and classication. And then applied this approaches to the thermograph y images. It identies the breast cancer b ut can only be suitable for database ha ving less thermograph y images. The accurac y of the result may A utomated br east cancer detection system fr om br east mammo gr am using deep ... (Suneetha Chittineni) Evaluation Warning : The document was created with Spire.PDF for Python.
582 ❒ ISSN: 2502-4752 3. B A CKGR OUND 3.1. Deep neural netw ork One of the sub di visions of machi ne learning is deep learning model. The deep learning is designed by including more hidden layers in the traditional neural netw orks. The hidden layers are present in-between the input layer and output layer . The deep neural netw ork (DNN) becomes popular i n medical eld because it pro vides high performance in e xtracting the features from the images [13]. In order to pro vide good perfor - mance DNN requires huge dataset for training the model. Selecting the h yper -parameter is also an important process in DNN to e xtract the optimal features from the mammographic image dataset. 3.2. Con v olutional neural netw ork There are numerous deep neural netw orks the most commonly used neural netw ork by the researches are con v olutional neural netw ork (CNN) [14]. Normally a CNN consists of a set of feed forw ard layers, this feed forw ard layers e x ecutes the con v olutional lter , pooling layer and fully connected layers that helps to e xtract the image features. By learning the input image patterns CNN allo w feature e xtraction this is done in feature e xtraction layer/con v olutional layer [15]. CNN is used for tuning the h yper parameters lik e batch size, number of epochs, acti v ation layer and learning rate to e xtract the features of mammograph y images. So, radiologists are not needed to se gment the breast cancer image features. 4. PR OPOSED METHODOLOGY In the proposed approach the chest mammographic images are obtained from the MIAS image database. The obtained images are pre-processed to mak e all the images in same size. Then the features are e xtracted from the pre-processed images by a con v olutional neural netw ork inte grated with rat sw arm optimization al- gorithm. This algorithm tunes the paramet ers by updating the location of the rat to e xtract the breast cancer features from the images. At last arithmetic optimization algorithm inte grated with random forest approach classies the normal and breast cancer af fected patients from the e xtracted images. The reasoning problem in CNN is eliminated by the arithmetic optimization algorithm. 4.1. Pr e-pr ocessing Pre-processing is applied to the mammograph y image database to eliminate the unw anted noise in- cluded in the images. By pre-processing the features that are needed for detecting the breast cancer are sharp- Indonesian J Elec Eng & Comp Sci, V ol. 25, No. 1, January 2022: 580–588 v ary based on the dimension of the lump and also produce f alse positi v e rate. Chen et al. [4] introduced an abbre viated protocol (AP) for M RI that identies the cancerous lumps in the bre ast. This protocol has tw o other protocols abbre viated protocol1 (AP1) and abbre viated protocol2 (AP2). Maximum intensity projection (MIP) and rst post-contrast subtracted (F AST) images were grouped together to form AP1 protocol. AP2 protocol w as a combination of AP1 protocol with dif fusion-weighted imaging (D WI). These tw o protocols e xamine the ultrasound images and then detect the breast cancer . But this model has the limitations that it doesn’ t consider the past history of the patients also the small lumps in the breast are not identied in this method. Aslam et al. [11] implemented an automatic deep con v olutional neural netw ork (DCNN) approach for identifying the breast cancer . This approach rst g at h e rs the data from tw o datasets. Then utilizes the con v olutional neural netw ork layers for training the data and then classies the breast cancer patients. The per - formance of this approach w as based on the number of data a v ailable for training. If the training data decreases the performance of this approach also decreases. Ibrahim et al. [12] uses thermal images to identify the breast cancer . The thermal ima g e s are g athered from the database for mastology research with infrared image (DMR- IR). The g athered thermal images under go pre-processing and se gmentation. After that the cancerous features are e xtracted from the se gmented image. Then the breast cancer patients were classied from the e xtracted images. During this process v arious algorithms were used in e v ery stage that may cause man y problems lik e setting the k-v alue and the data after mer ging totally changed from its original size and density , which produces wrong prediction. T o o v ercome the abo v e limitations the mammograph y images are used in this article. From the mammograph y im ages it is dif cult to identify the cancerous lumps for that well e xperienced e xperts are needed and the y ha v e to e xamine the mammograph y image clearly to detect the breast cancer correctly . All the time the e xperts are not a v ailable so to ease their w ork an automated system is implemented to detect the breast cancer using con v entional neural netw ork and arithmetic optimization algorithm. Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 583 ened and the image quality also impro v ed [16]. This process does not change the features of the original image it only enhances the features. The pre-processing uses 2D median image lter function that increases the mam- mograph y image quality by clearing and f ading the unw anted image portions out of sight and mak es the image suitable for further processing [17], [18]. The mechanism of this lter is it mo v es e v ery pix el one by one and then e v ery pix el v alue is altered by the median of neighbouring pix el v alue. 4.2. F eatur e extraction In this process, the cancerous features are e xtracted from the pre-processed mammographic images by tuning the h yperparameters using con v olutional neural netw ork inte grated with rat sw arm optimization (CNN- RSO) algorithm. The rat sw arm optimization algorithm is a bio inspired optimization algorithm that describes the public acti vit ies of rat and sw arm [19]. Here the rat is the predator that tries t o catch the sw arm which is the victim. This algorithm tunes the h yperparameters batch size, number of epochs, acti v ation layer , and learning rate by that it alters the location of the rat. The group of rats tries to hunt the sw arm by chasing and ghting with it. The predator chasing the victim is mathematically modelled in (1). The information about the locality of the victim is kno wn by the best search agent. Based on the location of the best search agent the other search agents can modify their locations. − → L = U . − → L ( x ) + V . ( − → L r ( x ) − − → L i ( x )) (1) Here, the location of the rat is represented by − → L i ( x ) and the ideal solution is represented by − → L r ( x ) . The v alues for the v ariables U and V are computed as, U = R − x ( R M ax I ter at ion ) (2) V = 2 .r and () (3) where, the v alues of x=0,1,2,. . . , M ax I ter ation . R and V are the random numbers that v aries from 1 to 5.The aggressi v e ghting of the rat with the sw arm to kill him is mathematically computed as follo ws: − → L i ( x + 1) = | − → L r ( x ) − − → L | (4) here, the ne xt modied location of the rat is represented by − → L i ( x + 1) . Each time the location of the rat changes the best ideal solution is stored in − → L i ( x + 1) . Algorithm 1 sho ws the h yperparameter tuning of rat sw arm optimization. Thus the h yperparameters batch size, number of epochs, acti v ation layer and learning rate are tuned to e xtract the breast cancer features from the pre-processed mammograph y image dataset. Algorithm 1 Hyperparameter tuning of rat sw arm optimization Input: the batch size − → L i (i=1,2,. . . ,n) Output: the best optimal e xtracted image dataset Procedure HyperparameterRSO Initialize the parameters U, V and R Compute the tness v alue for each image dataset − → L r ← the best image dataset while x < M ax I ter ation do f or eachimag edataset do Change the location of the present image dataset by (4) end f or Change the parameters U, V and R V erify if an y image dataset goes out of the gi v en image dataset then adjust it Compute the tness v alue for each image dataset Change Lr if an y better solution is found x = x+1 end while Return − → L r A utomated br east cancer detection system fr om br east mammo gr am using deep ... (Suneetha Chittineni) Evaluation Warning : The document was created with Spire.PDF for Python.
584 ❒ ISSN: 2502-4752 4.3. Classication The breast ca n c er patients are classied from the feature e xtracted dataset by arithmetic optimiza tion algorithm inte grated with random forest (A O A-RF) [20]. The A O A i s a population-based algorithm so the ideal solution cannot be found in a single step. It tak es much iteration to found the best ideal solution. The best ideal solution in A O A is obtained by the arithmetic operators addition (A), subtraction (S), multiplication (M), and di vision (D). Initialization phase, e xploratory phase and e xploitati v e phase are the three processes in the A O A approaches. 4.3.1. Initialization phase In initialization phase, the random forest (RF) algorithm is implemented to retrie v e the best obtained or the nearly optimum solution [21]. The set of candidate solutions (C) is generated from the decision trees (DT) each iteration the ideal candidate solution is treated as a best obtained or the nearly optimum solution. There are L image dataset in the decision tree, the candidate solution C is represented in (5), C = c ( S , θ i ) i = 1 , 2 , ..., L (5) here, ith decision tree is repres ented by ( S , θ i ) . The samples for training is S and the single tree gro wth is represented as θ i . 4.3.2. Exploration phase The operators multiplication (M) and di vision (D) are considered as the operators for e xploration. These tw o operat ors produce high decision v alues which helps the e xploration phase to search the near ideal solution. The e xploration phase can also be used in e xploitati v e phase to assist it to nd the accurate breast cancer patients. F or this process it applies tw o techniques: di vision (D) search approach and multiplication (M) search approach. This technique is represented in (6). x i,j ( P iter + 1) = ( best ( x j ) / ( M O P + ∈ ) ∗ (( U V j − LV j ) ∗ µ + LV j ) , r 2 < 0 . 5 best ( x j ) ∗ M O P ∗ (( U V j − LV j ) ∗ µ + LV j ) , O ther w ise (6) Where, r denotes the random number , UV and L V represents the upper v alue and lo wer v alue, the result of ith location in the ne xt iteration is x i,j ( P iter + 1) , P iter represents the present iteration. The math optimizer probability (MOP) is calculated in (7). The maximum number of iteration is represented as M ax iter . M O P ( P I ter ) = 1 − P 1 / ∞ I ter M ax 1 / ∞ I ter (7) 4.3.3. Exploitation phase The operators addition (A) and subtraction (S) are considered as the operators for e xploitation. These tw o operators produce lo w decision v alues which help the e xploitation phase to choose the best ideal dataset. The e xploitation phase is mathematically modelled as (8). x i,j ( P iter + 1) = ( best ( x j ) / M O P ∗ (( U V j − LV j ) ∗ µ + LV j ) , r 3 < 0 . 5 best ( x j ) ∗ M O P ∗ (( U V j − LV j ) ∗ µ + LV j ) , O ther w ise (8) This e xploitation phase is similar to the e xploration phase b ut it does not jammed in an y dataset while searching. The nal classied breast cancer patients are obtained by (9), B ( C ) = ar g max ( x i,j )( b r ,b,l ( C ) = i ) i = 1 , 2 , ..., N (9) where, B(C) represents the nal classied breast cancer patients, x i,j is the number of near ideal dataset. Indonesian J Elec Eng & Comp Sci, V ol. 25, No. 1, January 2022: 580–588 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 585 5. RESUL T AND DISCUSSION In MA TLAB R2018a softw are the proposed system is implemented. The Mammographic Image Analysis Society (MIAS) database consists of 322 breast mammograph y images is used in the proposed system for the e xperimentation purpose. In that 206 are normal images and 113 are breast cancer images [22]. The proposed approach is compared with other classiers lik e Nai v e Bayes (NB) [23], k-nearest neighbor (KNN) [24], decision tree (DT) [23], support v ector machine (SVM) [24] and random forest (RF) [25]. The performance metrics considered for e v aluation are ac curac y , sensiti vity , F1-Score, prec ision, specicity and Kappa statistic. Figure 2 sho ws the accurac y and precision for v arious algorithms. From that it is pro v ed that the proposed A O A-RF produce high accurac y compared to other approaches. Figure 3 sho ws the performance of F1-score and kappa for v arious algorithms. Figure 2. Performance of accurac y and precision Figure 3. Performance of F1-score and kappa Both the F1-score and kappa v alues are relati v ely high for the proposed approach. Fi gure 4 sho ws the performance of sensiti vity and specicity for v arious algorithms. From these gures it is clear that the perfor - mance of the proposed classier random forest inte grated with arithmetic optimization algorithm is superiorly high compared to other algorithms. Figure 4. Performance of sensiti vity and specicity The root mean square error (RMSE) and mean absolute error (MAE) are combined into one to detect the error in the breast cancer dataset. Figure 5 represents the RMSE and MAE error for the proposed A O A-RF and for v arious other e xisting algorithms such as DT , KNN, SVM, NB, RF . From that it is e vident that the proposed classier produces less error compared to other classiers. The proposed approach with and without rat sw arm optimization (RSO) v alues are sho wn in T able 1. Figure 6 sho ws the proposed approach performance with RSO and without RSO. A utomated br east cancer detection system fr om br east mammo gr am using deep ... (Suneetha Chittineni) Evaluation Warning : The document was created with Spire.PDF for Python.
586 ❒ ISSN: 2502-4752 Figure 5. Comparati v e analysis of the proposed A O A-RF and e xisting DT , KNN, SVM, NB, RF Figure 6. Performance of proposed approach with and without RSO T able 1. Proposed approach with and without RSO Proposed W ith RSO Proposed W ithout RSO P arameters V alues P arameters V alues Specicity 1 Specicity 0.7333 FPR (1-Specicity) 0 FPR (1-Specicity) 0.2667 TPR (Sensiti vity) 1 TPR (Sensiti vity) 0.7692 Error 0 Error 0.2500 Precision 1 Precision 0.7143 F-measure 1 F-measure 0.7407 Accurac y 1 Accurac y 0.7500 MCC 1 MCC 0.5013 Kappa 1 Kappa 0.5000 The proposed approach with RSO produces high accurac y , precision, sensiti vity , specicity , F1-score and kappa v alues nearly 1 compared to the proposed approach without RSO. The recei v er operating characteris- tics (R OC) curv e analysis of proposed classier with RSO and without RSO is sho wn in Figure 7. The recei v er operating characteristics (R OC) curv e analysis of proposed classier A O A-RF and other dif ferent classiers with CNN as feature e xtraction is sho wn in Figure 8. Figure 7. Proposed classier A O A-RF with and without RSO Figure 8. Comparison of proposed classier A O A-RF with other classiers The nal result obtained is sho wn by the confusion matrix. The confusion matrix obtained for A O A- RF classier without RSO is gi v en in Figure 9. The confusion matrix obtained for A O A-RF classier with RSO Indonesian J Elec Eng & Comp Sci, V ol. 25, No. 1, January 2022: 580–588 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 587 is gi v en in Figure 10. The proposed approach with RSO produces 100% accurac y which is relati v ely higher than proposed approach without RSO produces 75% accurac y . Figure 9. Confusion matrix for A O A-RF classier without RSO Figure 10. Confusion matrix for A O A-RF classier with RSO 6. CONCLUSION In this paper , no v el deep learning-based automatic breast cancer diagnosing systems from the mam - mographic images are de v eloped. This system helps the doctors/radiologists t o identify t he breast canc er au- tomatically . In this system, the mammographic images in MIAS dataset under goes image pre-processing by 2D median image lter to remo v e the noise in the dataset. The breast cancer features from the pre-processed mammographic images are retrie v ed using con v olutional neural netw ork inte grated with rat sw arm optimization (CNN-RSO) algorithm. Finally , the arithmetic optimization algorithm inte grated with random forest (A O A-RF) classier classies the breast cancer af fected and unaf fected patients. While analysing the performance of the A O A-RF classier with other classiers the performance of the proposed classier A O A-RF is relati v ely high. The proposed system A O A-RF with CNN-RSO produces 100% accurac y in detecting the breast cancer from the mammographic images. The system can further be impro v ed by increasing the size of the mammographic image dataset. REFERENCES [1] J. W ang, X. Y ang, H. Cai, W . T an, C. Jin, and L. Li, “Discrimination of Breast Cancer with Microcalcications on Mammograph y by Deep Learning, ” Scientic Reports , v ol. 6, p. 27327, 2016, doi:10.1038/srep27327. [2] C. Aroef, Y . Ri v an, and Z. Rustam, “Comparing random forest and support v ector machines for breast cancer classica- tion, ” TELK OMNIKA (T elecommunication, Computing, Electronics and Control) , v ol. 18, no. 2, pp. 815–821, 2020, doi: 10.12928/telk omnika.v18i2.14785. [3] M. A. Kah ya, “Classication enhancement of breast cancer histopathological image using penalized logistic re gression, ” Indonesian J ournal of Electrical Engineering and Computer Science (IJEECS) , v ol. 13, no. 1, pp. 405–410, 2019, doi: 10.11591/ijeecs.v13.i1.pp405-410. [4] S. Q. Chen, M. Huang, Y . Y . Shen, C. L. Liu, and C. X. Xu, “ Abbre viated MRI Protocols for Detecting Breast Cancer in W omen with Dense Breasts, ” K orean Journal of Radiology , v ol. 18, no. 3, pp. 470–475, 2017, doi: 10.3348/kjr .2017.18.3.470. [5] N. R. Sheno y and A. Jatti, “Ultrasound image se gmentation through deep learning based impro vised U-Net, ” Indonesian Journal of Electrical Engineering and Computer Science (IJEECS) , v ol. 21, no. 3, pp. 1424–1434, 2021, doi: 10.11591/ijeecs.v21.i3.pp1424- 1434. [6] M. S. Islam, N. Kaabouch and W . C. Hu, “ A surv e y of medical imaging techniques used for breast cancer detection, ” IEEE Interna- tional Conference on Electro-Information T echnology , EIT 2013 , 2013, pp. 1–5, doi: 10.1109/EIT .2013.6632694. [7] J.-L. Gonzalez-Hernandez, A. N. Recinella, S. G. Kandlikar , D. Dabydeen, L. Medeiros, and P . Phatak, “T echnology , application and potential of dynamic breast thermograph y for the detection of breast cancer , ” International Journal of Heat and Mass T ransfer , v ol. 131, pp. 558–573, 2019, doi: 10.1016/j.ijheatmasstransfer .2018.11.089. [8] M. H. Alhabib and O. H. Alhabib, “Detection of partially o v erlapped masses in mammograms, ” Indonesian Journal of Electrical Engineering and Computer Science (IJEECS) , v ol. 18, no. 1, pp. 235–241, 2020, doi: 10.11591/ijeecs.v18.i1.pp235-241. [9] Z. Cao, L. Duan, G. Y ang, T . Y ue, and Q. Chen, “ An e xperimental study on breast lesion detection & classication from ultrasound images using deep learning architectures, ” BMC Medical Imaging , v ol. 19, no. 51, 2019, doi: 10.1186/s12880-019-0349-x. [10] D. Singh and A. K. Singh, “Role of ima ge thermograph y in early breast cancer detection- P ast, present and future, ” Computer Methods and Programs in Biomedicine , v ol. 183, p. 105074, 2020, doi: 10.1016/j.cmpb .2019.105074. [11] M. A. Aslam, Aslam, and D. Cui, “Breast Cancer Classication using Deep Con v olutional Neural Netw ork, ” Journal of Ph ysics: Conference Series , v ol. 1584, p. 1584, 2020, doi: 10.1088/1742-6596/1584/1/012005. [12] A. Ibrahim, S. Mohammed, and H. A. Ali, “Breast Cancer Detection and Classication Using Thermograph y: A Re vie w , ” Adv ances in Intelligent Systems and Computing , v ol. 723, pp. 496–505, 2018, doi: 10.1007/978-3-319-74690-6 49. A utomated br east cancer detection system fr om br east mammo gr am using deep ... (Suneetha Chittineni) Evaluation Warning : The document was created with Spire.PDF for Python.
588 ❒ ISSN: 2502-4752 [13] W . Liu, Z. W ang, X. Liu, N. Zeng, Y . Liu, and F . E. Alsaadi, “ A surv e y of deep neural netw ork architectures and their applications, ” Neurocomputing , v ol. 234, pp. 11–26, 2017, doi: 10.1016/j.neucom.2016.12.038. [14] S . Z. Ramadan, “Using Con v olutional Neural Netw ork with Cheat Sheet and Data Augmentation to Detect Breast Cancer in Mam- mograms, ” Computational and Mathematical Methods in Medicine , v ol. 2020, pp. 1–9, 2020, doi: 10.1155/2020/9523404. [15] K. K umar and A. C. S. Rao, “Breast cancer classication of image using con v olutional neural netw ork, ” 2018 4th International Conference on Recent Adv ances in Information T echnology (RAIT) , 2018, pp. 1–6, doi: 10.1109/RAIT .2018.8389034. [16] A . K. Jothi and P . Mohan, “ A Comparison between KNN and SVM for Breast Cancer Diagnosis Using GLCM shape and LBP Features, ” 2020 Third International Conference on Sm art Systems and In v enti v e T echnology (ICSSIT) , 2020, pp. 1058–1062, doi: 10.1109/ICSSIT48917.2020.9214235. [17] W . Chen, P . Chen, Y . Hsiao and S. Lin, “ A Lo w-Cost Design of 2D Median Filter , ” IEEE Access , v ol. 7, pp. 150623–150629, 2019, doi: 10.1109/A CCESS.2019.2948020. [18] B. John and S. Nallathambi, “Study and analysis of lters, ” Adv ances in Computational Sciences and T echnology , v ol. 10, no. 3, pp. 331–341, 2017. [19] G. Dhiman, M. Gar g, A. Nag ar , V . K umar , and M. Dehghani, “ A no v el algorithm for global optimization: Rat Sw arm Optimizer , ” Journal of Ambient Intelligence and Humanized Computing , v ol. 12, pp. 8457–8482, 2021, doi: 10.1007/s12652-020-02580-0. [20] L. Ab ualig ah, A. Diabat, S. Mirjalili, M. A. Elaziz, and A. H. Gandomi, “The Arithmetic Optimization Algorithm, ” Computer Methods in Applied Mechanics and Engineering , v ol. 376, p. 113609, 2021, doi: 10.1016/j.cma.2020.113609. [21] F . Guo et al. , “Random F orest Algorithm-based Multi-Feature V ector Optimization for F atigue Dri ving V igilance Monitoring, ” in 2020 Chinese Control And Decision Conference (CCDC) , 2020, pp. 3113–3118, doi: 10.1109/CCDC49329.2020.9164300. [22] M. Mustra and A. Stajduhar , “Se gmentation Masks for the Mini-Mammographic Im age Analysis Society (mini-MIAS) Database, ” IEEE Consumer Electronics Mag azine , v ol. 9, no. 5, pp. 28–33, 1 Sept. 2020, doi: 10.1109/MCE.2020.2986799. [23] B. T . Ahmed, “Data mining techniques for lung and breast cancer diagnosis: A re vie w , ” International J ournal of Informatics and Communication T echnology (IJ-ICT) , v ol. 10, no. 2, pp. 93–103, 2021, doi: 10.11591/ijict.v10i2.pp93-103. [24] A. Agrima, I. Mounir , A. F archi, L. Elmaazouzi, and B. Mounir , “Emotion recognition from syllabic units using k-nearest-neighbor classication and ener gy distrib ution, ” International Journal of Electrical and Computer Engineering (IJECE) , v ol. 11, no. 6, pp. 5438–5449, 2021, doi: 10.11591/ijece.v11i6.pp5438-5449. [25] H. Bun yamin and T . T un ys, “ A Comparison of Retweet Prediction Approaches : The Superiority of Random F orest Learning Method, ” TELK OMNIKA (T elecommunication, Computing, Electronics and Control) , v ol. 14, no. 3, pp. 1052–1058, 2016, doi: 10.12928/TELK OMNIKA.v14i3.3150. BIOGRAPHIES OF A UTHORS Suneetha Chittineni is an Associate Professor in the Department of Computer Applica- tions at R.V .R. & J.C. Colle ge of Engineering, Guntur , India. She recei v ed a Ph.D. de gree in Com- puter Science and Engineering with specialization in Articial Intelligence from Acharya Nag arjuna Uni v ersity , Guntur , India. Her research interests include Articial Intelligence, Machine Learning, Deep Learning, and Data Mining. Dr . Suneetha has published more than 20 research papers in v arious international journals. She can be contacted at email: suneethachittineni@gmail.com. Sai Sandeep Edara is currently a B.T ech nal year student pursuing Computer Sci- ence & Engineering in R.V .R. & J.C. Colle ge of Engineering, Guntur , India. His areas of inter - est include Data Science, Machine Learning, and Deep Learning. He can be contacted at email: esaisandeep2001@gmail.com. Indonesian J Elec Eng & Comp Sci, V ol. 25, No. 1, January 2022: 580–588 Evaluation Warning : The document was created with Spire.PDF for Python.