Comparison of a Classifier Performance Testing Methods: Support Vector Machine Classifier on Mammogram Images Classification

Abstract

This paper compares between testing performance methods of classifier algorithm on a standard database of mammogram images. Mammographic interchange society dataset (MIAS) is used in this work. For classifying these images tumors a multiclass support vector machine (SVM) classifier is used. Evaluating this classifier accuracy for classifying the mammogram tumors into the malignant, benign or normal case is done using two evaluating classifier methods that are a hold-out method and one of the cross-validation methods. Then selecting the better test method depending on the obtained classifier accuracy and the running time consumed with each method. The classifier accuracy, training time and the classification time are considered for comparison purpose.

Breast Cancer Detection in Mammogram Images Using K–Means++ Clustering Based on Cuckoo Search Optimization

“…A mammogram image used in this study is drawn from different datasets, some of which are from the Mini–MIAS [ 5 , 9 , 15 , 16 , 17 , 18 , 19 ] and DDSM [ 12 ], while others are from BCDR [ 10 ]. We used the same images to evaluate their detection methods [ 3 , 4 , 5 , 6 , 9 , 10 , 12 ].…”

“…BCDR contains 18 normal and 105 abnormal images from 123 images. Before discussing the mammogram image, it may be helpful to describe the structure of the image; see Figure 2 [ 5 , 9 , 10 , 12 , 15 , 16 , 17 , 18 , 19 ]. Figure 2 a is an image from Mini–MIAS, Figure 2 b is an image from INbreast, and Figure 2 c is an image from BCDR, respectively.…”

“…BCDR contains 18 normal and 105 abnormal images from 123 im ages. Before discussing the mammogram image, it may be helpful to describe the structu of the image; see Figure 2 [5,9,10,12,15,17,18]. Figure 2a…”

“…In this stage, binary image segmentation using the Gaussian OTSU's threshold [34] is used. Given an image: P(i) is the normalized frequency of i, and is calculated according to Equation (15).…”

“…Anji et al [11] DNNS MG Cancer --97.21 Azlan et al [12] PCA and SVM DDSM and Mini-MIAS 93.94 96.61 95.24 George et al [13] The hybrid methods by combined the CNN and SVM BreaKHis 97.24 96.18 96.21 Alanazi et al [14] CNN architectures Kaggle 162 H&E --87.00 Mohammed et al [15] SVM Mini-MIAS --91.12 Agnes et al [16] MA-CNN Mini-MIAS 96.00 --Yektaei et al [17] Multiscale Convolutional Neural Network (MCNN) Mini-MIAS 95.90 -97.03 Kaur et al [18] Multiclass Support Vector Machine (MSVM) Mini-MIAS --96.90 Viswanath et al [19] SVM Raw sample images -68 84.84 Jin et al [20] Binary classifier with CNNI-BCC Mini-MIAS --73.24 Kayode et al [ From a comparative review of machine learning and performance parameters used for evaluating the detection method, each method is suitable for detecting breast cancer on different datasets. Nevertheless, the accuracy of each technique still varies for different datasets.…”