Diagnosis of colorectal cancer based on imperialist competitive algorithm

Abstract

The Colorectal cancer leads to more number of death in recent years. The diagnosis of Colorectal cancer as early is safe to treat the patient. To identify and treat this type of cancer, Colonoscopy is applied commonly. The feature selection based methods are proposed which helps to choose the subset variables and to attain better prediction. An Imperialist Competitive Algorithm (ICA) is proposed which helps to select features in identification of colon cancer and its treatment. Also K-Nearest Neighbor (KNN) classifier is used to retain a minimal Euclidean distance between the feature of query vector and all the data in the nature of prototype training. Experimental results have proved that the proposed method is superior when compared to other methods in its metrics of performance. Better accuracy is achieved by the proposed method.

Keywords

Colorectal cancer (CRC)Colonoscopy feature selection imperialist competitive algorithm (ICA)k-Nearest Neighbor (k-NN) classifier

1 Introduction

A research body that is increasing has shown that the microenvironment of cancer is very different and complex to the extent that they can be seen as immunologic organs that are distinct. Colorectal Cancer (CRC) as mentioned is one of the primary reasons for deaths related to cancer [1]. Many systematic methods for pathological condition diagnosis have led to a high rate of early detection of CRC patients thereby reducing mortality rates. For diagnosing pathological conditions might contribute to a high detection rate of patients at early stages of CRC, leading to a reduction in mortality rates. Sigmoidoscopy and a blood test known as fecal occult test are currently the methods of screening that has reduced mortality due to CRC [2, 3]. But these also have their own limitations; the fecal occult test of blood has low sensitivity and flexible sigmoidoscopy on the other hand is not a comfortable technique for patients. 19-9 (CA19-9) a carbohydrate antigen and Carcino Embryonic Antigen (CEA) are widely used as markers for detecting various types of cancer like liver, stomach, pancreatic, colon etc., But these markers have low sensitivity for detection mainly when the disease is at its early stages. So diagnostic markers that are CRC specific markers have to be developed and sensitive, non-invasive and rapid screening have to be ensured.

Patients suffering from Inflammatory Bowel Diseases (IBD), Crohn’s disease (CD), Ulcerative Colitis (UC) have a higher rate of risk of CRC [4]. 18% was the magnitude of risk at 30 years post the initial diagnosis that was in UC3. As a consequence of this the guidelines for a professional society make recommendations that colonoscopies for surveillance have to ideally start between 8 and 10 years post diagnosis with a repeat after every 2 or 3 years on people with colonic CD or UC. In the literature that is recent there are conflicts based on secular changes pertaining to risks of IBD or with CRC patients. Most of the studies that come from cohorts based on population show that IBD and CRC incidence has reduced. Many reasons exist for this kind of a temporal reduction like wide usage of maintenance and its treatment, increase in frequency of colectomy, increase in uptake in surveillance colonoscopy in practice in allowing identification of dysplastic lesions before CRC develops.

In the general population, screening Colonoscopy screening in generic population has been connected with a reduction in the mortality rate of colorectal cancer. Those with negative colonoscopy are under a risk of further lesions especially on the period of follow up that can extend up to 10 years after the first examination. However, data that can prove whether this kind of protective effect really exists among IBD [5] patients that have not undergone examination. In fact, this is a critical clinical question as the molecular pathogenesis of cancer is different from sporadic carcinoma. Therefore, the surveillance colonoscopies and their benefits can never by extrapolated by comparing the normal population with IBD patients. Further, some more factors can affect the potential estimates. The decline in the risk of CRC patients that have IBD can also bring down the benefits of regular examinations of surveillance. There is also an existing controversy that the best method to perform the surveillance examinations as well as the yield of biopsies that are randomly taken. The individuals that are diagnosed with IBD when their age is very less have to examine colonoscopies change the risk of the resultant patient population after CRC [6].

Mucosa visualization of the large intestine and the distal terminal ileum can normally be made possible in colonoscopy. Colon cancer incidence can be reduced by removal of Polyps. Colonoscopy is a very popular method for the evolution of the colon in many adults that have symptoms of large bowel, results that are abnormal when a radiographic study is made, deficiency of iron or anemia, CRC screening positive results, post polypectomy, surveillance of post cancer resection and surveillance after diagnosis in cases of bowel diseases that are anti-inflammatory.

When a set of these features are used a classification can be performed by any algorithm of machine learning. In the years of the past machine learning applications or recognition of patterns and their domain have expanded manifold. Many new techniques have been included to address the issue of reduction in variables that are not relevant or redundant which can hinder the performance of various tasks. Selection of feature through elimination can be of great help in understanding data, bringing down requirement of computation as well as the problems in dimensionality and improving of performance of predictors [7]. The methods of selection of features can be through three different models which are 1. Filter method 2. Wrapper method and 3. Embedded method. The first approach designs an independent measure for a given algorithm. So the features that precisely indicate the original data are used. But this method does not consider the inter feature interactions. In this approach the following methods based on correlation for selection of feature are present [8], the t-test, mutual information, information gain, and finally methods based on entropy. The models known as Wrapper models normally concentrated on the improvisation of the accuracy of classification of problems of patterns and their recognition in order to bring about filter models that are better. But the wrapper approach is quite lengthy. The embedded techniques make used of an algorithm that is inductive representing the classifier as well as the feature selector.

Classification [9] is perhaps the most common problems that is taken up for study in Statistics, databases and machine learning. Many such algorithms have been proposed in the past, but recently cancer classification has been gaining prominence among researchers. Different machine learning systems that are stochastic are being applied for the treatment of cancer. They are hierarchical clustering. Support Vector Machines (SVM), k-nearest neighbors, self-organizing maps and Bayesian networks.

Some ideal benefits in using K-Nearest Neighbor (KNN) is robust training data to noisy ones which is effective in case of large data and very fast in case of small training sets. This does not need knowledge of the structure of data that is contained in the training set or retraining along with the training present when new patterns of training are included. The better prediction of Colorectal Cancer is done by proposed method.

2 Related work

Rad et al. [10] made a selection of a feature set best suited for classifying the varieties of rice on the basis of images of huge samples by means of an algorithm called imperialist competition. This was a new as well as an evolutionary method of optimization which derived its inspiration form competition in imperialistic situations. Results proved that feature sets chosen by this method gave better performance of classification in comparison to genetic technique of algorithm.

Dienstmann et al. [11] made a focus on molecular, gene and pathologic expression of characterizations in the early stages of cancer of colon; there were new insights into the biomarkers that made predictions and prognostication insights thereby helping in the definition of proper adjuvant treatments for patients in clinical practice of routine nature.

Moraru et al. [12] further presented another different approach to analyze colon cancer in ultrasonography. A method known as speckle suppression was presented. A wavelet transform known as Daubechies was used for the shift invariance and its property and additional information on the plane of the complex wavelet domain in comparison with the real domain. This method gave good results and proved the image usefulness and their techniques of processing the diagnosis of imaging. The ROI had a variance known as local echogenicity which was used for the comparison of the distribution of the local echogenicity inside the image acquired. This image was further analyzed with histograms that could explain images of gray-level. This is a very valuable and important information to help discriminate tumors.

Ananthakrishnan et al. [13] analyzed data from a total of 6823 people with IBD in which 2764 had colonoscopy recently and 4059 without. They were followed for 3 years in hospitals in Boston. The result of this was a CRC diagnosis. An examination of the number of patients that underwent colonoscopy from within 36 months before diagnosis or also when a follow up period ended was made. They excluded ones performed within 6 months before diagnosis. A process Multivariate logistic regression was done. Colonoscopy (within 36 months) was connected with a low incidence of patients with CRC also with IBD and a much lower rate of mortality of those with CRC.

Sears & Garrett [14] discussed the insights of micro biota contributions to explore a variety of framework for the evaluation of the role of microbes in causing cancer. Some findings were pointed out on CRC-potentiating species and knowledge gaps. Lastly, the microbial metabolism and its role was examined in relation to xenobiotics, bile acids and diet in both etiology and also in therapeutics of the CRC.

Siegel et al. [15] made a provision for the statistics of cancer of colon which made an inclusion of the most recent data on incidence of this cancer, rate of survival and mortality. The first data was procured from the Epidemiology of the National Cancer Institute’s Surveillance. The final result and the Epidemiology and finally the central cancer registry of the Association of North America. The death rates of Colorectal cancer has come down by 2% in the 90’s and further to 3% recently. The progress made in bringing this down and the decrease in death rates can further be brought down by improving access and usage of screening as well as standard treatment.

Dehal et al. [16] further examined the connection between T2DM or Type2 Diabetes Mellitus and the rate of survival among patients that also had CRC and analyze if this varies according to insulin treatment, duration of insulin use and sed. Out of 2278 people diagnosed with rectal cancer or colon cancer from 1992 to 2007 a study of incidence of cancer was made. Those who had CRC along with T2DM had a very high mortality risk than those with only CRC without T2DM.

Vieira et al., [17] proposed a Modified Binary Particle Swarm Optimization (MBPSO) method for feature selection with the simultaneous optimization of Support Vector Machine (SVM) kernel parameter setting, applied to mortality prediction in septic patients. An enhanced version of Binary Particle Swarm Optimization (BPSO), designed to cope with premature convergence of the BPSO algorithm is proposed. MBPSO control the swarm variability using the velocity and the similarity between best swarm solutions. This work uses SVMs in a wrapper approach, where the kernel parameters are optimized at the same time. The approach is applied to predict the outcome (survived or deceased) of patients with septic shock. Further, MBPSO is tested in several benchmark datasets and is compared with other PSO based algorithms and GA. The experimental results showed that the proposed approach can correctly select the discriminating input features and also achieve high classification accuracy, especially when compared to other PSO based algorithms. When compared to GA, MBPSO is similar in terms of accuracy, but the subset solutions have less selected features.

Tsai et al. [18] studied the vital statistical features with Gray Level Co-occurrence Matrix (GLCM), Discrete Wavelet Transform (DWT), Spatial filters, Wiener filter, Gabor filters, Haralick features, fractal filters, and Local Binary Pattern (LBP) for colorectal cancer tissue identification by Support Vector Machine (SVM) and decision fusion of feature selection. The average experimental results achieve high identification rate which is significantly superior to the existing known methods. The proposed method outperforms and attains a high classification accuracy rate at 93.17% for eight classes and 96.02% for ten classes which validate promising applications for cancer tissue classification of histological images.

Zhao et al. [19] make use of an ensemble model to classify samples into healthy and CRC groups and enhance prediction performance. The genes are selected by Minimum Redundancy Maximum Relevance (mRMR) technique to decrease the dimensionality of features and to enhance the prediction results. The problem caused by imbalanced data using hybrid sampling algorithm RUSBoost is solved. The Whale Optimization Algorithm (WOA) is applied to find most optimal parameters of the proposed MKF-SVM. The results show that the proposed model attains a higher G-means compared to other methods.

3 The establishment of model

3.1 Colonoscopy database with polyps from Cornell University

In 2004 when presenting the NCI Executive Committee the ACRIN proposal to conduct the National CT Colonography Trial (6664), a case was made that publicly accessible image data sharing would offer a valuable research asset to a wide image processing research community. Adding to the many merits of that proposal, the data-sharing component was strongly endorsed. ACRIN completed the trial expeditiously and its results were published in NEJM in fall 2008 to wide interest. ACRIN has graciously allowed the wider research community access to a portion of the data from that trial here on TCIA, including spreadsheets identifying positive and negative polyp cases. The complete ACRIN 6664 protocol description can be found at http://www.acrin.org/TabID/151/Default.aspx.

The Figs. 1 to 3 shows the sample images for colonoscopy database with polyps.

Fig.1

Sample Image1 for colonoscopy database with polyps.

Fig.2

Sample Image2 for colonoscopy database with polyps.

Fig.3

Sample Image3 for colonoscopy database with polyps.

3.2 Daubechies wavelet transform

The wavelet transforms of Daubechies haves been defined just like the wavelet transform of Haar was done by calculating the averages of running and its difference through scalar products along with signals for scaling them the difference between them only being how the signals of scaling are actually defined. This type of wavelet has a very balanced response to frequency but they are non-linear. These wavelets make use of overlapping windows, to ensure that he coefficient of high frequency can reflect all changes in frequency. This makes the Daubechies wavelets more useful in the removal of noise and compression when processing audio signals [20].

3.3 Gabor filter

Texture features are extracted by Gabor filters in the retrieval of medical images. These filters are used for identification and detection of image or texture. This can extract features that are edge like in different scales and also orientations where different locations of images are observed. These features resemble those that are procured by the visual system of humans. More specifically, texture segmentation, analysis and recognition is done by 2D Gabor filters. They are sinusoidal functions that are complex that are handled by a rotated Gaussian [21]. The function of the Gabor is shown in the spatial domain Equation (1): $g (x, y) = g_{a} (x^{'}, y^{'}) . e^{(2 π jfx)}$ (1)

In which the rotated co-ordinates (x′, y′) are $(x^{'}, y^{'}) = (x cos φ + y sin φ, - x sin φ + y cos φ)$ (2)

And g_a denotes the Gaussian function.

For the extraction of textural features an image that is textured into an input and later transformed through a bank of Gabor filters and then a Gaussian smoothing filter in cascades. A feature vector can be constructed from the images and their statistics of outputs that comes from the filter bank branches. In case a filter bank consisting of N number of branches is taken for extraction of features a Gabor filter denoted by G_i(i = 1,2, ... ,N) which is complex, the filter output also contains numbers that are complex. Therefore in case of a magnitude operator that is non-linear, the modulus of all the complex numbers at the point of output is computed that is at a post Gaussian stage is P_i (i = 1, 2, ... , N) which is the Gaussian post filter (I_i) and the output required is the needed is the feature image, where the output statistics provides texture features. Let W_i(j,k) denote (2m+1)x(2n+1) in the neighbourhood of (j,k)’th pixel of the output image I_i. Feature vector for (j,k)’th pixel (given by F_jk, needed for classification pixel wise) of an input image I which is formed in statistics of second order statistics of the local pixel features (j,k). The construction of feature vector and of its length 2xN is given in Equation (3): $\begin{matrix} F_{jk} (2 r - 1) = μ [W_{r} (j, k)] and \\ F_{jk} (2 r) = σ [W_{r} (j, k)]; \\ r = 1, 2, . ., N \end{matrix}$ (3)

In which, μ (·) and σ (·) denote the operation of computing mean and standard deviation.

3.4 Genetic algorithm (GA) based feature selection (FS)

It comprises of an optimal feature selection search implemented by GA. The feature sets are specified to classifiers and the performance of classifiers for each subset is examined. Genetic algorithm, a population based optimization method that efforts to mimic a natural evolution process of man. It trails an iterative process to produce new population from the parent population through cross over and mutation. The fitness solutions of candidates are assessed by fitness function which depends on the objective of optimization process.

3.5 Imperialist competitive algorithm

Imperialist Competitive Algorithm (ICA), which is derived from social evolution of humans becomes a component of the theory swarm intelligence. ICA, is a meta-heuristic technique which imitates the social behavior of the imperialistic competition between the countries in solving problems of optimization. The initial version of the algorithm has been used in solving optimization problems continuously [22]. In its similarity to other algorithms that are evolutionary ICA with its first random population known as countries. Every one of this has an n-dimensional problem of optimization which is denoted as an 1 × n array as shown by Equation (4): $Country = [p_{1}, p_{2}, . . . p_{n}]$ (4)

From among those countries that possess ideal function values of cost imperialists are chosen. The first few empires are formed by colony distribution between imperialists based in proportion on the power that has been normalized by the imperialists (p_n) as depicted in Equation (5): $p_{n} = | \frac{c_{n}}{\sum_{1}^{n} c_{i}} |$ (5)

In which c_n is cost of nth imperialist, where the imperialist’s normalized cost (C_n) is defined by means of Equation (6): $C_{n} = c_{n} - max (c_{i})$ (6)

Once the distribution of colonies between imperialists is completed closing of the empires corresponding to them takes place. The result of this is that the entire power of the empire, is nothing but the imperialist country’s power along with an additional power percentage of their respective colonies. Now the empires get into the process of increasing their power by means of bringing closer to them the other empires. And the result of this is that there is a slow decrease in the weak empire’s power thereby an increase in the stronger empire’s power takes place automatically. The manner in which the strongest empire competes with the weakest colony belonging to the weakest of the empires is known as imperialistic competition. As the power of an empire increases it begins to attract more and more weak colonies thereby rendering the weak empires unable to see success in their competition with others and not able to improve the condition of their power which slowly gets eliminated completely based on some algorithm iteration. The final result of this is the totality of power of all the strong empires gradually goes up and weak empires completely collapse. After this takes place the colonies belonging to that empire is distributed to the remaining empires. Finally, all the colonies gradually become one state belonging to one empire for the entire world and making all the other countries its colonies [23].

Algorithm for ICA:

Initialization of the Empires

Moving of the colonies to the imperialist relevant to them

Verifying the presence of a colony where the cost is higher than the imperialist

Conduct an exchange between the imperialist and the colony positions

Otherwise total cost of empires to be computed

Picking of the colony that is the weakest from among the weakest empire and hand it over to the empire that is most likely to process it

Check if there is an empire without colonies

Elimination of the empire without a colony

Otherwise stop and check if condition is satisfied

Completed. If not re start from step 2.

Figure 4 shows the flowchart of Imperialist Competitive Algorithm.

Fig.4

Flowchart of imperialist competitive algorithm.

For selection of features, the solutions has to represent a set of feature subset. When the ICA algorithm starts, empires are initialized. This represents set of solutions, thus, this set is made up of features. It is initialized with a random set of feature subset. On iterating through the algorithm, the optimal subset is obtained.

3.6 K-Nearest Neighbor (KNN) Algorithm

A method for the classification of objects on the basis of training examples that are closest to the feature space is known as k-nearest neighbor algorithm [24]. k- NN is that learning that is instance based also known as lazy learning in which the functions are locally approximated and until classification is complete all types of computation are deferred. One of the simplest types of algorithms of machine learning is the k-nearest neighbor algorithm which was originally proposed by Cover and Heart in the year 1968. This is an object that is identified by the neighbors’ majority of votes by an object that is assigned to the most common class from among its neighbors that are k-nearest where k is a small integer that is positive. When k = 1, the assigning of the object is made to the commonest value that is the nearest examples of training to k. All ties are broken randomly. The nearest neighbor algorithm of k that has been used in classification of neighborhood as a value of prediction of the value of the query instance that is new. From among a set of those objects the neighbors are pulled out where the classification is correctly available. Normally any object is classified in terms of the distance that is has from that of its neighbors that is with the objective of ideally being assigned to the relevant class that is the commonest among the nearest neighbors of k distance. The key word for this algorithm is distance. Every single object here is duly represented by vectors of a feature space that is multidimensional where the Euclidean distance can be used for the purpose of calculating the actual distance that exists between both the positions of the vectors.

Classification of k Nearest Neighbor takes place on finding a group of k objects by the algorithm in a set of training that is actually the closest in terms of input and the classification of the input which is based on the class and its majority in that group. The prime elements that are needed for the approach are 1. A labeled objects set 2. Metrics and number of k’s distance. The algorithm for k-nearest neighbor and its classification is as in Fig. 5.

Fig.5

Algorithm for K-nearest neighbor (KNN).

KNN algorithm is chosen because of its advantages such as: it is applied to the data from any distribution, very simple, flexible classification scheme.

3.7 Naïve Bayes (NB)

A naive bayes classifier is based on a bayes theorem based classifier of naïve bayes [25] is capable of getting an efficient performance in classification. The probabilities of condition are P (x_j|c_i) and the probabilities prior to it are P (c_i). P (c_i) which is calculated by the counting of the samples of training and then by the dividing of the result of the count by using the set size of training. The naive bayes classifier is shown in Equation (7): $F_{i} (x) = P (x_{j} | c_{i}) P (c_{i})$ (7)

Table 1

Probability of Miss Detection

	GA based feature	GA Based feature	Imperialist competitive	Imperialist competitive
	selection,	selection	algorithm	algorithm
	Naïve Bayes	- KNN	- Naïve Bayes	- KNN
Classification Accuracy	0.9545	0.9425	0.9715	0.9677
Positive Predictive Value for Positive sentiment	0.9831	0.981	0.9845	0.9826
Positive Predictive Value for Negative sentiment	0.9141	0.8903	0.9519	0.9453
Sensitivity for Positive sentiment	0.9417	0.9238	0.9686	0.9641
Sensitivity for Negative sentiment	0.9746	0.9718	0.9761	0.9732
F measure for defect	0.9434	0.9293	0.9638	0.959
F measure for no defect	0.962	0.9515	0.9765	0.9733

In which X = (x₁, x₂, … x_N) denotes the feature vector, c_j denotes class labels, j = 1, 2, ... , N. This model is a simpler version of the model known as Bayesian probability theory. The naive Bayes classifier is on the assumption that features do not depend on it but operate on it. One attribute’s probability does not affect that of the other. It performs well on the classification but however it shows that errors happen owing to three factors which are bias, noise in training data and variance. The noise in training data can be reduced only by choosing good data for training. This bias has a huge impact on the variance which is small while training data is grouped.

4 Results and discussion

The experiments are carried out using Mathlab tool. Table 1 shows the summary of results for Classification accuracy, positive predictive for positive and negative sentiment, sensitivity for positive and negative sentiment and F measure for defect and no defects.

4.1 Accuracy

Considers both true positives and true negatives over all instances. In other words, accuracy shows the ratio of all correctly classified instances. $Accuracy = \frac{TP + TN}{TP + FP + TN + FN}$ (8)

Where: TP: True Positive, also called a hit

TN: True Negative, also called a correct rejection

FP: False Positive: also called a false alarm or a Type I error

FN: False Negative, also called a miss or a Type II error

Precision, also called Positive Predictive Value (PPV) is the fraction of retrieved instances that are relevant to the problem. $PPV = \frac{TP}{TP + FP}$ (9)

A perfect precision score (PPV = 1.0) means that every result retrieved by a search was relevant, but says nothing about whether all relevant documents were retrieved.

Sensitivity, which indicates the fraction of positives correctly classified as such, computed as: $Sensitivity = \frac{TP}{TP + FN}$ (10)

Where, True Positive (TP), True Negative (TN), False Positive (FP) and False Negative (FN).

F-measure: F-measure is a harmonic mean of precision and recall. $F - Measure = \frac{2 \times (Precision \times Recall)}{Precision + Recall}$ (11)

Where $Recall = \frac{TP}{TP + FN}$ (12)

Figs. 5 to 8 shows the same.

It is observed from Table 1 and Fig. 6 that the classification accuracy of imperialist competitive algorithm with naïve bayes performs better by 1.77% than GA based feature selection with Naïve Bayes, by 3.03% than GA based feature selection with KNN and by 0.39% than imperialist competitive algorithm with KNN.

Fig.6

Classification accuracy.

It is observed from Table 1 and Fig. 7 that the positive predictive value for positive sentiment of imperialist competitive algorithm with naïve bayes performs better by 0.14% than GA based feature selection with Naïve Bayes, by 0.36% than GA based feature selection with KNN and by 0.19% than imperialist competitive algorithm with KNN. Over all positive predictive value for positive sentiment performs better than negative sentiments.

Fig.7

Positive predictive value for positive and negative sentiment.

It is observed from Table 1 and Fig. 8 that the sensitivity for negative sentiment of imperialist competitive algorithm with naïve bayes performs better by 0.15% than GA based feature selection with Naïve Bayes, by 0.44% than GA based feature selection with KNN and by 0.29% than imperialist competitive algorithm with KNN. Over all sensitivity for negative sentiment performs better than negative sentiments.

Fig.8

Sensitivity for positive and negative sentiment.

It is observed from Table 1 and Fig. 9 that the F measure for no defect of imperialist competitive algorithm with naïve bayes performs better by 1.49% than GA based feature selection with Naïve Bayes, by 2.59% than GA based feature selection with KNN and by 0.33% than imperialist competitive algorithm with KNN. Overall F measure for no defect performs better than with defect.

Fig.9

F measure for defect and no defect.

In [17], the classification accuracy for colon database on 94.26 for BPSO & 95.84 for IBPSO, in our work the classification accuracy for ICA-naive bayes on 97.15.

5 Conclusion

Colorectal Cancer is a very common one which caused huge malignancies and health issues all the world over. The mortality from this has come down in the last twenty years as detection is early and good treatments are available. A general searching method known as ICA from the imperialistic competition from the interactions that are geo-political between countries. This algorithm is extremely sensitive to the data’s local structure. This one of the algorithms which have proved to be extremely simple and easy to practice. Results have proved that the accuracy of the classification of the competitive accuracy of the imperialist which has the naïve bayes that can perform much better 1.77% than that of a feature selection based on GA with the KNN and further by 0.39% in comparison with the imperialist algorithm with the KNN. The predictive values that are positive its sensitivity and the proposed method’s F measures that can give better performance.

References

Ratnaraju

, Swamy

G.N

, PadmaRaju

, Wrapper based Feature Selection for Virtual Colonoscopy Classification, Indian Journal of Science and Technology 9(S1) (2016), 1–8.

Wolf

A.M

, Fontham

E.T

, Church

T.R

, Flowers

C.R

, Guerra

C.E

, LaMonte...

S.J

, Walter

L.C

, Colorectal cancer screening for average-risk adults: 2018 guideline update from the American Cancer Society, CA: A Cancer Journal for Clinicians 68(4) (2018), 250–281.

Ogata-Kawata

, Izumiya

, Kurioka

, Honma

, Yamada

, Furuta...

, Watanabe

, Circulating exosomal microRNAs as biomarkers of colon cancer, PloS one 9(4) (2014), e92921. pp. (1–9).

Andersen

, Halfvarson

, Vogel

, Colorectal cancer in patients with inflammatory bowel disease: can we predict risk?, World J Gastroenterol 18(31) (2012), 4091–4094.

Sebastian

, Hernandez

, Myrelid

, et al., Colorectal cancer in inflammatory bowel disease: Results of the 3rd ECCO pathogenesis scientific workshop (I), J Crohns Colitis 8 (2013), 5–18.

Jess

, Simonsen

, Jorgensen

K.T

, et al., Decreasing risk of colorectal cancer in patients with inflammatory bowel disease over 30 years, Gastroenterology (2012), 375–381.

, Sun

, Gao

, Xu

, An ensemble feature selection technique for cancer recognition, Bio-medical Materials and Engineering 24(1) (2014), 1001–1008.

Chandrashekar

, Sahin

, A survey on feature selection methods, Computers & Electrical Engineering 40(1) (2014), 16–28.

Archetti

, Castelli

, Giordani

, Vanneschi

, Classification of colon tumor tissues using genetic programming, Artificial Life and Evolutionary Computation 49 (2010), 1–10.

10.

Rad

S.M

, Tab

F.A

, Mollazade

, Application of imperialist competitive algorithm for feature selection: A case study on bulk rice classification, International Journal of Computer Applications 40(16) (2012), 41–48.

11.

Dienstmann

, Salazar

, Tabernero

, Personalizing colon cancer adjuvant therapy: selecting optimal treatments for individual patients, Journal of Clinical Oncology 33(16) (2015), 1787–1796.

12.

Moraru

, Moldovanu

, Nicolae

M.C

, De-Noising Ultrasound Images of Colon Tumors Using Daubechies Wavelet Transform, In Bunoiu

& Malaescu

(Eds.), AIP Conference Proceedings 1387(1) (2011), 294–299. AIP.

13.

Ananthakrishnan

A.N

, Cagan

, Cai

, Gainer

V.S

, Shaw

S.Y

, Churchill...

, Liao

K.P

, Colonoscopy is associated with a reduced risk for colon cancer and mortality in patients with inflammatory bowel diseases, Clinical Gastroenterology and Hepatology 13(2) (2015), 322–329.

14.

Sears

C.L

, Garrett

W.S

, Microbes, microbiota, and colon cancer, Cell Host & Microbe 15(3) (2014), 317–328.

15.

Siegel

, DeSantis

, Jemal

, Colorectal cancer statistics, CA: A Cancer Journal for Clinicians 64(2) (2014), 104–117.

16.

Dehal

A.N

, Newton

C.C

, Jacobs

E.J

, Patel

A.V

, Gapstur

S.M

, Campbell

P.T

, Impact of diabetes mellitus and insulin use on survival after colorectal cancer diagnosis: the Cancer Prevention Study-II Nutrition Cohort, Journal of Clinical Oncology 30(1) (2011), 53–59.

17.

Vieira

S.M

, Mendonça

L.F

, Farinha

G.J

and Sousa

J.M

, Modified binary PSO for feature selection using SVM applied to mortality prediction of septic patients, Applied Soft Computing 13(8) (2013), 3494–3504.

18.

Tsai

M.J

, Yuadi

, Tao

Y.H

, Colorectal Cancer Tissue Classification Based on Machine Learning.

19.

Zhao

, Liu

, Zheng

, He

, Lu

, Lyu

, Whale optimized mixed kernel function of support vector machine for colorectal cancer diagnosis, Journal of Biomedical Informatics 92 (2019), 103124;.

20.

Elfouly

F.H

, Mahmoud

M.I

, Dessouky

M.I

, Deyab

, Comparison between haar and daubechies wavelet transformions on fpga technology, International Journal of Computing 6(3) (2014), 23–29.

21.

Naghdy

, Ros

, Todd

, Norahmawati

, Cervical cancer classification using Gabor filters, In Healthcare Informatics, Imaging and Systems Biology (HISB), 2011 First IEEE International Conference on (2011 Jully) (pp. 48–52). IEEE.

22.

Aleardi

, Pierini

Comparing Firefly Algorithm, Imperialist Competitive Algorithm and Quantum Particle Swarm Optimization on analytic objective functions and residual statics corrections, In 37 convegno del gruppo nazionale di geofisica della terra solida (2018), (pp. 172–176).

23.

Ebrahimi

, Mollazade

, Babaei

Toward an automatic wheat purity measuring device: A machine vision-based neural networks-assisted imperialist competitive algorithm approach, Measurement 55 (2014)196–205.

24.

Halder

, (2013).An Approach to Diagnosis of Cancer using k-nearest Neighbor (k-NN) Algorithm (Doctoral dissertation, JADAVPUR UNIVERSITY KOLKATA).

25.

Keerthika

, Priya

D.S.

Feature subset evaluation and classification using naive Bayes classifier, Journal of Network Communications and Emerging Technologies (JNCET) 1(1) (2015)22–27.