INTRODUCTION:

Cancer remains a most important cause of mortality in the world. In 2018, about the overall estimation of 1,735,350 cancer cases are getting more than 4,700 new cancer diagnoses each day [21]. The probability of lifetime being diagnosed with tumor is 39.7% for men and 37.6% for women which is a slightly more than 1 in 3. Early detection is one of the most hopeful approaches to reduce the growth of cancer. It knows that over 80% of malignancies come about in epithelial exteriors, most of which can be directly envisioned [22]. Consequently, many current techniques for tumor screening begin with visual examination of the whole tissue at risk in white light illumination, after that biopsy of highly suspicious tissue areas.

Biopsy is an incursive technique which reasons patient distress and it is suffered from sampling errors. Noninvasive is one of alternative have been required using a number of medical imaging modalities including magnetic resonance imaging (MRI), computed tomography (CT) and ultrasound, Optical medical images may offer a possible solution to need for affordable imaging tools to help in early identification and supervision of cancer.

Classification is an easy task, but then it is being challengeable to the machine. The image classification is included image pre-processing, image sensors, object detection, object segmentation, feature extraction and object classification. The classification system involves of a database which has predefined patterns that compare with an object to classify [14]. Image Classification is a significant task in several fields such as biomedical images, biometry, robot navigation and remote sensing. In supervised classification, the some pixels are known categories and provides the label to classes. This method is known as training. Then classifier practices trained pixels for classify other images. In unsupervised classification, pixels are categories with the aid of their properties. This method is known as clustering. In this user select how many groups or clusters is he needs. The unsupervised classification used while not trained pixels are obtainable. In image classification several methods are used: Artificial Neural Network (ANN), Support Vector Machine (SVM) and Decision Tree.

RELATED WORKS:

Rahul Kumar Sevakula, and Nishchal Kumar Verma et al [1] presented Classification algorithms which reduce errors caused by variance as well as by bias. But, there take place several conditions like that low generalization error, minor overfitting in the test results. This system was used majority vote point (MVP) classifiers, because of very low Vapnik–Chervonenkis (VC) dimension, can show a generalization error which is even less than that of linear classifiers. In conclusion of this system, case studies on prostate tumor detection problem and machine fault diagnosis problems revalidate by MVP classifier that achieve a lower generalization error. In MVP classifier has many problems that lacks enough flexibility to fit the training data; this paper suggested that DL-based feature transformations can be thoroughly investigated in the future.

Yunxiang Mao, Zhaozheng Yin and Joseph Schober et al [2] presented the number of Circulating Tumor Cells (CTCs) in blood shows the tumor comeback to chemotherapeutic agents and disease evolution. In this proposed system a Deep Convolutional Neural Network (DCNN) with regularly learned features for image-based finding of CTC and also training samples to outline the boundary of classification amid positive and negative samples. In this proposed systems performance of DCNN on a challenging problem in dataset that it is hopeful to resolve the automated problem CTC detection in a non-invasive technique. This paper detection approach could be improved to microfluidics devices for perfect and quick enumeration of CTCs in clinical blood samples for timely diagnosis and treatment observing.

Amir Zjajo, Rene van Leuken et al Robust et al [3] presented dynamic and efficient spike classifier that capable of perfect identification of the neural spikes in low SNR. In the proposed system used spike classifier, scalable, programmable neural based on spike detection of nonlinear energy operator, and a multiclass kernel support vector machine (SVM) classification. Clustering spike derivative features is, because of the infecting noise, a challenging big task, the degree of overlay among the annotated clusters growths as a purpose of the noise variance. The ability to differentiate spikes from noise, and to discriminate spikes from different causes from the superimposed waveform, so be subject to on both the inconsistencies between the noiseless spikes from each source.

Ming-Chi Wu,Wen-Chi Chin and Ting-Chen Tsan et al [4] proposed system used the Unsharp Mask to enhance the image edge with the aim of decrease the computation time and increase the efficiency and specify an ellipse region of interest (ROIs) of nasopharynx in the image. Later, this presented system used histogram equalization to denoise and use Nobuyuki Otsu methods to get the threshold value to binaries it in that region. At that moment the proposed system used texture and geometric feature extraction method to detect the tumor, and MRI image with Neural-Fuzzy based Ada Boost classifier to diagnose malignant or benign tumors of the nasopharynx. This methods used to reduce the misjudgment rate.

Zhan-Li Sun, Chun-Hou Zheng, Qing-Wei Gao, Jun Zhang, and De-Xiang Zhang et al [5] presented Eigengene removed by independent component analysis (ICA) is one type of effective feature for tumour classification. This proposed system used Eigengene and support vector machine (SVM) based classifier committee learning (CCL) algorithm which to improve the diversity of weaker classifiers. Gene expression data built by several feature subspaces are showed by ICA. In addition, Bayesian sum rule (BSR) is used to incorporate the outputs of the weaker SVM classifier. Tentative results on DNA microarray datasets validate to classify. One difficulty of this approach is training time of the CCL algorithm increases significantly. The training time has decreased by reducing the number of weaker SVM classifiers. But, it is a problem to design an effective criteria to choice the weaker SVM classifiers.

K. S. Thara, K. Jasmine, et al [6] presented tumour of Brain detection is the most significant way to define the early tumor. Expanding the tumor is being a challenge task because of the complex characteristics of the MRI Images that provides high intensive, divergent and an ambiguous boundaries. In the proposed system input MRI Image is pre-processed, followed by K Means clustering method and Fuzzy C Means clustering method segmentation is used and the classification has done by using the Fuzzy probabilistic neural network classifier (FPNNC) that used to classify the MRI Image as malignant or benign. This proposed method gives better result compared to the existing classifiers.

V. Amsaveni, N. Albert Singh and J. Dheeba et al [7] In this paper present a new classification method using Cascaded Correlation Neural Network (CCNN) is a nonlinear classifier for detection tumour of brain from MRI. This paper presented Gabor texture features are occupied from the image Region of interest (ROI). The proposed method several images collected from diagnostic centers used with real time images. Using the real MRI image database an accuracy score was greatly done using the proposed approach.

Wei Luo, Lipo Wang and Jingjing Sun et al [8] presented the support vector machine (SVM) for cancer classification and a mixed two-step feature selection method. The firstly used a modified t-test method to choice discriminatory features and secondly removes principal components from the genes depend on the modified t-test method. This paper is tested two-step method in three data sets likewise 1.lymphoma data set, 2.SRBCT data set and 3. ovarian cancer data set. For example, for the leukemia data set, the biggest score gene is gene X95735.

S. Khazendar,H. Al-Assam, H. Du, S. Jassim ,A. Sayasneh, T. Bourne and J. Kaijser, D. Timmerman et al [9] proposed a novel method for automatic ovarian tumour classification established on decision level fusion which first extracts two different types of features likewise Histogram and Local Binary Pattern (LBP) from ovary of ultrasound images. Support Vector Machine (SVM) classifier used to classify ovarian tumor based on features separately. Experimental results on several ultrasound images of ovarian tumour provide classification accuracy of results based on classification of benign and different types of malignant tumours decisions of high, medium and low confidence respectively. This paper argues that such confidence depended prediction results are more meaningful than others.

Hemita Pathak Vrushali Kulkarni et al [10] proposed system which acquires ultrasound images, image processing and machine learning algorithms perfectly classify normal and abnormal tumours in ovarian cancer. This method denoise ultrasound image using wavelet transform, grey level texture features removed using

Grey Level Co-occurrence Algorithm (GLCM), extracted features trained through Support vector machine (SVM) and chosen non-redundant features selected through Relief-F and test through SVM for output. Proposed method was validated by several normal and abnormal tumours of ultrasound image. Hence, the features are significant for outcome and initial results show that the proposed system can be more reliable for ovarian tumor classification and also this is fully automated.

Xiangyong Cao, Feng Zhou, Lin Xu, Deyu Meng, Zongben Xu, and John Paisley et al [11] presented a supervised classification algorithm for Hyper-Spectral Image (HSI) which integrates spatial and spectral data in a unified Bayesian framework. First, formulated the HSI classification problem from a Bayesian outlook. Then, implemented a convolutional neural network (CNN) in combination with a Markov random field to classify HSI to acquire the posterior class distributions. Following, spatial data is considered by employing a spatial smoothness prior on the labels. Lastly, this system iteratively updated the CNN parameters using stochastic gradient decent and update the class labels of all pixel vectors. Compared with the other the classification system achieves better performance on one artificial data set.

Sarni Suhaila Rahim,Vasile Palade ,James Shuttleworth and Chrisina Jayne et al [12] presented an automatic detection of diabetic maculopathy and retinopathy in eye fundus images by using fuzzy techniques. The proposed system is consisted of four steps, in detail: image acquisition, image pre-processing with four retinal structures localization, feature extraction and classification of image. This system are applied a combination of fuzzy image processing methods, the Circular Hough Transform (CHT) and some feature extraction techniques. The proposed system also presented a new technique for the macula region localization with the aim of detect the maculopathy. Besides the proposed detection system presented to diagnose process through online dataset and the dataset collection which compared to other public eye fundus image databases such as the UCI machine learning repository.

Margarita Osadchy, Daniel Keren, Dolev Raviv et al [13] presented a problem in computer vision is group recognition. In the propose system is used “hybrid” classifier which is implemented by linear classifier and kernel classifier to alleviate this difficulty. Hybrid classifier replaced the negative samples by a prior and discovers a hyperplane which split up the positive samples by prior. This technique is extended to an ensemble-based approach and to kernel space. This resulting binary classifiers achieved better classification rate, reduced training complexity, compared to SVM.

Robert Pike,Guolan Lu and Dongsheng Wang et al [14] presented system is to improve a classification technique that syndicates both spectral and spatial data from hyperspectral images. This proposed automated algorithm is established on a minimum spanning forest (MSF) and optimal band choice which is proposed to classify accurate normal and abnormal tumour tissue boundary on hyperspectral images. This system used a support vector machine (SVM) classifier which is trained to make a pixel-wise classification probability map of normal and abnormal tumour tissue. This probability map is used to find markers that are used to work out mutual data for a range of bands in the hyperspectral image and thus select the optimal bands. Minimum Spanning Forest (MSF) is lastly developed to segment the image using spatial and spectral information.

Chuan-Yu Chang, Hui-Ya Hu and Yuh-Shyan Tsai et al [15] presented a prostate cancer detection using dynamic MRIs. Initially, an ACM (Active Contour Model) is trained and used to segment the prostate MRI image and then, some features are extracted from the dynamic MRIs when injection at different time and converted them into RIC curves. Next, some discriminative features are designated by Fisher’s Discrimination Ration (FDR) and Sequential Forward Floating Selection (SFFS). Lastly, the support vector machine (SVM) classifier is implemented to classify the segmention of prostate cancer into two kinds: malignant and benign.

Fabio A. Spanhol , Luiz S. Oliveira, Caroline Petitjean, and Laurent Heutte et al [16] proposed system introduced a dataset of several breast cancer histopathology images called BreaKHis picked up on patients. The dataset contains both benign tumour and malignant tumour images. This work is related with this dataset that is automated classification of these images in two types, which is a valuable computer-aided diagnosis (CAD) tool for the clinician and another one providing this dataset and an identical evaluation protocol.

Argin Margoosian and Jamshid Abouei et al [17] proposed two cancer classification methods based on multicategory microarray data sets. Because of the high dimensionality of microarray data sets, selecting reliable feature selection and classification procedures with a higher amount of accuracy and a low complexity is a vital task in bioinformatics. This Proposed system is used two consistent ensemble-based classifier techniques specifically the ensemble of k-nearest neighbor (ENN) and the ensemble of naive bayes (ENB). Simulation results of classifiers have significantly better accuracy than Support Vector Machine (SVM) and artificial neural networks in the field of multicategory microarray tumour classification based on cancer data set. To decrease the run time complexity though maintaining the same classification accuracy as before, this system used the recursive feature elimination depend on the multiple support vector machine (SVM) classifier to choice more informative genes In advance put on the ensemble-based classifiers.

Fengying Xie, Haidi Fan, Yang Li, Zhiguo Jiang, Rusong Meng, and Alan Bovik et al [18] presented classifying melanocytic tumors as normal or abnormal by the study of digital dermoscopy images. This system follows three steps: initially, lesions are takeout using a self-generating neural network (SGNN) and then, features descriptive of cancer color, texture and border are pull out; and last one, lesion matters are classified using a classifier based on a neural network ensemble model (NNEM). In this proposed system model is designed a network ensemble classifier (NE) that combined back propagation (BP) neural networks with fuzzy neural networks (FNN). Experimentations are carried out on two various dermoscopy databases that contain images of both the Caucasian and xanthous races.

T. M. ShahriarSazzad, L. J. Armstrongand A. K. Tripathy et al [19] presented ultrasound image technique used to identify bigger and more mature tissues rather than small ovarian tissues. The proposed system processes digitized color histopathology images picked up from biopsy slide and classified ovarian reproductive tissues automatically.

N. Varuna Shree and T. N. R. Kumar et al [20] presented noise removal method with extraction of gray-level co-occurrence matrix (GLCM) features, Discrete Wavelet Transform (DWT) based on brain tumor area developing segmentation to decrease the complexity and improve the performance. This proposed system was followed by morphological filtering which eliminates the noise that can be made after segmentation. The system is used probabilistic neural network (PNN) classifier and trained the data and also tested accuracy of brain tumour detection from MRI image. This tentative results achieved great accuracy in classifying benign and malignant tissues from brain MRI images.

SUMMARY OF TUMOUR DETECTION AND CLASSIFICATION WORK:

Image processing techniques, MRIs, CT, Ultrasound, PET and X-Ray images and also image dataset together have exposed significant development in medical prediction and decision making cancer of tumour. The expert doctors diagnose the disease and find the stage of cancer by practice. The treatment consist of radiation therapy, surgery, targeted therapy and chemotherapy which are painful, long and costly. Therefore, an effort is made to automated classifier used to detect the tumour using image processing techniques in easy way.

In the MVP classifier has many problems, which deficiencies adequate flexibility to fit the training data. The high performance of Deep Convolutional Neural Network (DCNN) on a challenging in dataset. In eigengene and SVM based CCL algorithm training time of the CCL algorithm rises significantly. The training time is decreased by decreasing the number of weaker classifiers. Though, it was still a big problem to design a real criteria to pick the weaker classifiers. In ovarian tumours based on decision level fusion feature extracted to classify the some malignant tumour mistakenly using LBPH. In ovarian mass through ultrasound Images using MLT this proposed system was feature extraction GLCM by GLRLM than lower accuracy. In recognition using hybrid classifiers larger memory requirements in SVM training. In a dataset for breast cancer histopathological image classification proposed systems had problems in multiclass classification for both the normal and the abnormal image sets.

For prospect work increasing the number of images used for the process and will improve better results of the Accuracy, PSNR, Sensitivity, Specificity, MSE. Also Ultrasound, CT, MRI, X-ray and PET images can be considered for this automated classification technique. Comparison will be done for all these images. Consequently one can be justified that types of images gives better result for tumour detection.

CONCLUSION:

It is analyzed for different classifier and Image processing techniques which can be employed in automated tumour of cancer detection system. Now days various existing techniques have been used for the prediction of tumour of cancer at early stage. In this paper a summary of different technique for classification and image processing employed in the arena of cancer prediction. The highest concentration is on using various classifiers combined with several segmentation algorithms for detection of tumour using image processing. The summary of numerous classification and segmentation techniques with their classifier accuracy, selectivity, and sensitivity of detection of tumour has presented. Since the study it has established better result.

REFERENCES:

1. Rahul Kumar Sevakula, and Nishchal Kumar Verma. Assessing generalization ability of Majority Vote Point Classifiers, IEEE Transactions on Neural Networks and Learning Systems, 2017;28(12):1-13

2. Yunxiang Mao, Zhaozheng Yin and Joseph Schober. A Deep Convolutional Neural Network Trained on Representative Samples for Circulating Tumor Cell Detection, IEEE Winter Conference on Applications of Computer Vision (WACV), 2016:1-6

3. Amir Zjajo, Rene van Leuken, A 41 μW Real-Time Adaptive Neural Spike Classifier, IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI), 2016; 6(1):489-492

4. Ming-Chi Wu,Wen-Chi Chin and Ting-Chen Tsan. The Benign and Malignant Recognition System of Nasopharynx in MRI Image with Neural-Fuzzy based Adaboost Classifier, IEEE International Conference on Information Management (ICIM), 2016;5(4):1070-1074

5. Zhan-Li Sun, Chun-Hou Zheng, Qing-Wei Gao, Jun Zhang, and De-Xiang Zhang.Tumor. Classification Using Eigengene-Based Classifier Committee Learning Algorithm, IEEE signal processing letters, 2012;19(8):445-448.

6. K.S.Thara, K.Jasmine,Brain. Tumour Detection in MRI Images using PNN and GRNN, IEEE Conference International Conference on Wireless Communications, Signal Processing and Networking (WiSPNET), 2016;l.6.(16):1504-1510.

7. Amsaveni.V, Albert Singh.N and Dheeba. J. Computer aided detection of tumor in MRI Brain images using cascaded correlation Neural network, IET Chennai Fourth International Conference on Sustainable Energy and Intelligent Systems (SEISCON), 2013:12(4);527-532.

8. Wei Luo, Lipo Wang and Jingjing Sun. Feature Selection for Cancer Classification Based on Support Vector Machine, IEEE WRI Global Congress on Intelligent Systems,2009;5(9):422-426.

9. Khazendar,H.S Al-Assam,H. Du, S. Jassim ,A. Sayasneh, T. Bourne and J. Kaijser, and D. Timmerman. Automated Classification of Static Ultrasound Images of Ovarian Tumours Based on Decision Level Fusion, IEEE Computer Science and Electronic Engineering Conference (CEEC), 2014:9; 148-153.

10. Hemita Pathak Vrushali Kulkarni. Identification of Ovarian mass through Ultrasound Images using Machine Learning Techniques, IEEE International Conference on Research in Computational Intelligence and Communication Networks (ICRCICN), 2015:l.6;137-140.

11. Xiangyong Cao , Feng Zhou , Lin Xu, Deyu Meng ,Zongben Xu, and John Paisley. Hyperspectral Image Classification With Markov Random Fields and a Convolutional Neural Network, IEEE Transactions on Image Processing, 2018:27(5); 137-140.

12. Sarni Suhaila Rahim,Vasile Palade ,James Shuttleworth and Chrisina Jayne. Automatic screening and classification of diabetic retinopathy and maculopathy using fuzzy image processing, Springer Brain Informatics journal, 2016:3( 4); 249–267.

13. Margarita Osadchy, Daniel Keren and Dolev Raviv. Recognition Using Hybrid Classifiers, IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016:38(4); 1-30.

14. Robert Pike,Guolan Lu and Dongsheng Wang. A Minimum Spanning Forest-Based Method for Noninvasive Cancer Detection with Hyperspectral Imaging, IEEE Transactions on Biomedical Engineering, 2016:63(3); 1-11.

15. Chuan-Yu Chang, Hui-Ya Hu and Yuh-Shyan Tsai. Prostate Cancer Detection in Dynamic MRIs, IEEE International Conference on Digital Signal Processing (DSP), 2015:1(1); 1079-1282.

16. Fabio A. Spanhol, Luiz S. Oliveira, Caroline Petitjean, and Laurent Heutte. A Dataset for Breast Cancer Histopathological Image Classification,IEEE Transactions on Biomedical Engineering, 2016:63(7);1455-1462.

17. Argin Margoosian and Jamshid Abouei. Ensemble-based Classifiers for Cancer Classification Using Human Tumor Microarray Data,IEEE Iranian Conference on Electrical Engineering (ICEE), 2013:l(3);1-6.

18. Fengying Xie, Haidi Fan, Yang Li, Zhiguo Jiang, Rusong Meng, and Alan Bovik. Melanoma Classification on Dermoscopy Images Using a Neural Network Ensemble Model, IEEE Transactions on Medical Imaging, 2017: 36(3);1-6.

19. Shahriar Sazzad T.M, Armstrongand A L.J,Tripathy,K. An automated approach to detect human ovarian tissues using type P63 counter stained histopathology digitized color images, IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI),2016:1;25-28.

20. Varuna Shree.N and T. N. R. Kumar. Identification and classification of brain tumor MRI images with feature extraction using DWT and probabilistic neural network, Springer Brain Informatics journal,,2018:5(1);23–30

21. Ulagamuthalvi.V, Sridharan.D. Development of Diagnostic Classifier for Ultrasound Liver Lesion Images, International Journal of Computer Applications, 2012:.52(18); 12-15.

22. Noah Bedard,Mark Pierce, Adel El-Naggar, Anandasabapathy S, Ann Gillenwater and Richards-Kortum.R. Emerging roles for multimodal optical imaging in early cancer detection: a global challenge, Technology in Cancer Research and Treatment, 2010:9(2);1-6.

Received on 04.08.2018 Modified on 31.08.2018

Research J. Pharm. and Tech 2019; 12(1): 407-411.

DOI: 10.5958/0974-360X.2019.00074.X