Detection and Classification of Leukemia using MPFCM Segmentation and Random Forest with Boosting Techniques

: Identification of blood disorders is through visual inspection of microscopic blood cell images. From the identification of blood disorders lead to classification of certain diseases related to blood. We propose an automatic segmentation method for segmenting White blood cell images. Firstly, modified possibilistic fuzzy c-means algorithm is proposed to detect the contours in the image. The GLCM features are extracted and features are selected by MRMR. Adaptive boosting and LS Boosting has been utilized to classify blast cells from normal lymphocyte cells. Comparison performance of classification accuracy was carried out. The effectiveness of the classification system is tested with the total of 80 samples collected. The evaluated results demonstrate that our method outperformed the existing systems with an accuracy of 88 %.


Introduction
The assessment of the white blood cells in the bone marrow of patients is very informative in clinical practice. Segmentation is the process of partitioning a digital image into multiple segments based on pixels [7]. Segmentation is a critical and essential component of image analysis system [8]. The result of image segmentation is a collection of segments which combine to form the entire image. Various techniques have been proposed for segmenting an image in a better way. From the segmentation result, geometrical features such as area, perimeter etc [1] were detected for the final detection of immature cells. Three different classification techniques such as Tree Bagger, LS Boosting and ADA boosting were employed for classification [4] [5] [6], in order to classify the lymphocyte (WBC) as healthy and leukemic.

Proposed Methodology 2.1 MPFCM Clustering Segmentation
Image segmentation is the process of partitioning a digital image into multiple segments. Image segmentation is typically used to locate objects and boundaries in images [1].MPFCM is a good clustering algorithm [2] [3] to perform classification tests because it possesses capabilities to give more importance to typicalities or membership values. In order to avoid the constraint corresponding to the sum of all typicality values of all data to a cluster must be equal to one cause problems particularly for a big data set. It produces memberships and possibilities simultaneously, with the usual point prototypes or cluster centers for each cluster. The objective function is defined by Subject to the constraints =1 for all k and 0≤ ,≤1. Here a>0, b>0, m >0, and >0. U is the partition matrix. T is the typicality matrix. V is a vector of cluster centers, X is a set of all data points, z represents a data point, n is the number of data points and c is the number of cluster centers which are described by s coordinates.

2.1.1Algorithm
Step 1: Initialize prototype Step 2: For each cluster compute penalty parameter Step 3: For each prototype calculate the distance C , Step 4: Calculate membership and typicality values , 1 ≤ i ≤ C, I S S N 2 3 2 1 -8 0 7 X V o l u m e 1 3 N u m b e r 1 J o u r n a l o f A d v a n c e s i n c h e m i s t r y 5934 | P a g e J a n u a r y 2017 w w w . c i r w o r l d . c o m (6) Step 5: Update the values of prototypes.
Input CLL image Input CML image PFCM Output PFCM Output

Feature Extraction
Feature extraction is a special form of dimensionality reduction [1]. When the input data to an algorithm is too large to be processed, then the input data will be transformed into a reduced representation set of features. Textural features based on the gray level co-occurence matrix (GLCM) are extracted from each image that are used to distinguish between normal and abnormal cancer cells. Co-occurrence matrices are calculated for four directions: 0º, 45º, 90º and 135º degrees.
GLCM has following features: Autocorrelation, Contrast, Correlation, Cluster Prominence, Cluster Shade, Dissimilarity, Energy, Entropy, Homogeneity, Maximum probability , Sum of squares, Sum average, Sum variance, Sum entropy, Difference variance, Difference entropy, Information measure of correlation, information measure of correlation, Inverse difference normalized are listed in table.2 I S S N 2 3 2 1 -8 0 7 X V o l u m e 1 3 N u m b e r 1 J o u r n a l o f A d v a n c e s i n c h e m i s t r y 5935 | P a g e J a n u a r y 2017 w w w . c i r w o r l d . c o m

Feature Selection
Feature selection is the process of selecting a subset of relevant features for use in model construction. Features are selected based on MRMR and tabulated in table.3.

Minimum-Redundancy and Maximum-Relevance (MRMR)
Feature-selection method that can use either mutual information, correlation, distance/similarity scores to select features [10].
The redundancy of all features in the set S is the average value of all mutual information values between the feature xi and the feature xj : The MRMR criterion is a combination of two measures given above and is defined as follows Suppose that there are n full-set features. Let fi be the set membership indicator function for feature xi, so that fi=1 indicates presence and fi=0 indicates absence of the feature xi in the globally optimal feature set. Let ci=I(xi;c) and aij=I(xi;xj).
The above may then be written as an optimization problem I S S N 2 3 2 1 -8 0 7 X V o l u m e 1 3 N u m b e r 1 J o u r n a l o f A d v a n c e s i n c h e m i s t r y 5936 | P a g e J a n u a r y 2017 w w w . c i r w o r l d . c o m

Classification
Classification is the task of assigning to the unknown test vector, a label from one of the known classes. Classification methods aimed to find mathematical models to recognize the membership of each object to its proper class on the basis of a set of measurements [7]. Once a classification model has been obtained [8], the membership of unknown objects to one of the defined classes can be predicted.

LS Boosting
 It fits regression ensembles  At every step, the ensemble fits a new learner to the difference between the observed response and the aggregated prediction of all learners grown previously.  The ensemble fits every new learner to Y n -η f (x n), Where, Y n -observed response, f (x n) -aggregated prediction from all grown weak learners, η -learning rate(0-1)

ADA Boosting
ADA Boost is a machine learning algorithm [4] that boosts the performance of other learning algorithms, known as weak learners, by weighting and combining them. The basic idea is that multiple weak learners can be combined to generate a more accurate ensemble, known as a strong learner. Various versions of the ADA Boost algorithm [6] have proven to be very competitive in terms of prediction accuracy in a variety of applications.

Performance Measures
The final results of the ensemble classifiers were analyzed and its performance evaluation is done based on the results obtained from the confusion matrix. Also the results are compared based on the three parameters namely accuracy, specificity and sensitivity.

Confusion Matrix
A confusion matrix is a specific

Performance Parameters
The parameters are namely (i) Accuracy -statistical measure of how well a binary classification test correctly identifies or excludes a condition.

LS Boosting
Input -CLL Image Input -CML Image Classification Output Classification Output

Performance Comparison
The accuracy, specificity and sensitivity of adaptive and LS boosting is tabulated in table.4 comparison is done as shown in fig.4 where LS boosting has high performance I S S N 2 3 2 1 -8 0 7 X V o l u m e 1 3 N u m b e r 1 J o u r n a l o f A d v a n c e s i n c h e m i s t r y 5939 | P a g e J a n u a r y 2017 w w w . c i r w o r l d . c o m

5.Conclusion
Thus the modified possibilistic fuzzy c-means algorithm approach was evaluated and tested for various blood cells. This algorithm gives the segmentation result of white blood cells. This algorithm begins with detecting the cells in the region. By using those regions, white blood cells alone segmented. Then classification algorithms such as adaptive boosting and LS boosting were implemented and performance is measured.