Medical image processing promises major advances in medicine as higher fidelity images are produced, e.g., filters improve image quality to facilitate diagnosis or reduce radiation absorption. We evaluated the fuzzy peer group averaging technique in terms of effectiveness and stability. Effectiveness was measured by applying the filter to a set of mammograms and comparing the results to the obtained by different methods. This technique obtained the best peak signal-to-noise ratio values. Stability means that a filter applied to a mammogram must perform adequately in any case regardless of the type of tissue, the class of abnormality, and the severity. Thus, this part of the study focused on validating the stability of the fuzzy peer group averaging technique for its use in mammograms by demonstrating its effectiveness regardless the case. The normal distribution of the peak signal-to-noise ratio in the frequency histograms obtained validated this assumption.

Fuzzy PGA filter technique performance on the mini-MIAS database of mammograms

Now mammography can be defined as the most reliable method for early breast cancer detection. The main goal of this study is to design a classifier model to help radiologists to provide a second view to diagnose mammograms. In the proposed system medium filter and binary image with a global threshold have been applied for removing the noise and small artifacts in the pre-processing stage. Secondly, in the segmentation phase, a hybrid bounding box and region growing (HBBRG) algorithm are utilizing to remove pectoral muscles, and then a geometric method has been applied to cut the largest possible square that can be obtained from a mammogram which represents the ROI. In the features extraction phase three method was used to prepare texture features to be a suitable introduction to the classification process are first order (statistical features), local binary patterns (LBP), and gray-level co-occurrence matrix (GLCM), finally, SVM has been applied in two-level to classify mammogram images in the first level to normal or abnormal, and then the classification of abnormal once in the second level to the benign or malignant image. The system was tested on the MAIS the Mammogram image analysis Society (MIAS) database, in addition to the image from the Teaching Oncology Hospital, Medical City in Baghdad, where the results showed achieving an accuracy of 95.454% for the first level and 97.260% for the second level, also, the results of applying the proposed system to the MIAS database alone were achieving an accuracy of 93.105% for the first level and 94.59 for the second level.

/pdf/classification-of-mammograms-based-on-features-extraction-455ept95g3.pdf

Classification of mammograms based on features extraction techniques using support vector machine

Advances in medical imaging technologies have given rise for effective diagnostic procedures. The acquisition promptness and resolution enhancements of imaging modalities have given physicians more information, less invasively about their patients. Active contours are used to segment, match and track images of an atomic structure by manipulating constraints derived from the image data together with prior knowledge about the location, size, and shape of these structures. The level set method is referred as a part of active contour family. The major disadvantages of level set method are initialization of controlling parameters and time complexity. The proposed method adopts Robust Spatial Kernel Fuzzy C-Means (RSKFCM) and Lattice Boltzmann Method (LBM) to overcome these drawbacks. RSKFCM is based on standard Fuzzy C-Means algorithm which uses Gaussian RBF kernel function as distance metric and incorporates spatial information. The LBM uses the energy function to determine and reduce the actual processing time which addresses the time complexity. The proposed system combines both RSKFCM and LBM to form a hybrid approach, and the system is tested on a large set of MRI brain images and the experimental results are found to be improved with respect to time complexity.

http://ijaas.iaescore.com/index.php/IJAAS/article/download/9463/7849

Brain Image Segmentation using Level Set : An Hybrid Approach

Cancer of the breast is one of the world's most prevalent causes of death for women. Early and efficient identification is important for can care choices and reducing mortality. Mammography is the most effective early breast cancer detection process. Radiologists cannot however make a detailed and reliable assessment of mammograms due to fatigue or poor image quality. The main aim of this work is to establish a new approach to help radiologists identify anomalies and improve diagnostic precision. The proposed method has been applied through the implementation of preprocessing then segmentation of the images to get the region of interest that was used to find a texture features that were calculated based on first Order (statistical features), Gray-Level Co-Occurrence Matrix (GLCM), and Local Binary Patterns LBP (LBP). In the features selection phase mutual information (MI) algorithm is applied to choose from the extracted features collection suitable features. Finally, Multilayer Perceptron has been applied in two stages to classify the mammography images first to normal or abnormal, and secondly, classification of abnormal images into benign or malignant images. This method was implemented and gave an accuracy of 92.91 % for the first level and 93.15% for the second level classification.

/pdf/classification-the-mammograms-based-on-hybrid-features-2d750kanaa.pdf

Classification the Mammograms Based on Hybrid Features Extraction Techniques Using Multilayer Perceptron Classifier

Computational time is high for Multilayer perceptron (MLP) trained with back propagation learning algorithm (BP) also the complexity of the network increases with the number of layers and number of nodes in layers. In contrast to MLP, functional link artificial neural network (FLANN) has less architectural complexity, easier to train, and gives better result in the classification problems. The paper proposed an evolutionary functional link artificial neural network (EFLANN) using genetic algorithm (GA) by eliminating features having little or no predictive information. Particle swarm optimization (PSO) is used as learning tool for solving the problem of classification in data mining.  EFLANN overcomes the non-linearity nature of problems by using the functionally expanded selected features, which is commonly encountered in single layer neural networks. The model is empirically compared to MLP, FLANN gradient descent learning algorithm, Radial Basis Function (RBF) and Hybrid Functional Link Neural Network (HFLANN) . The results proved that the proposed model outperforms the other models.

/pdf/feature-selection-using-evolutionary-functional-link-neural-1pxvsfoxyd.pdf

Feature Selection using Evolutionary Functional Link Neural Network for Classification

Breast cancer is one of the most common causes of death among women globally. Accurate and early detection is necessary for decreasing mortality and increase treatment success rates. Mammogram image is currently one of the best ways to detect breast cancer in the early stages, but it contains many artifacts such as noise, labels, and pectoral muscles, that must be deleted or suppressed because it greatly affects the results of the diagnosis in the coming stages. Removing the pectorals muscle is the biggest problem because it possesses an intensity tissue that closely resembles the tissue of fat, glands, and tumors in the form of mammograms. In this paper, an effective algorithm has been suggested by Hybridization Bounding Box and Region growing algorithm (HBBRG) algorithm to solve the problem of pectoral muscle removal which greatly affects the results of tumor detection in the next stages by combines the Bounding Box (BB) and Region growing (RG). To perform this work, pre-processing for mammogram images was applied in two stages. In the first stage, a medium filter and binary image with a specific threshold were used to remove noise and label respectively. In the second phase, the pectoral muscles were removed by applying the (BB) and (RG) algorithm separately, and then we proposed merging the two methods to set up an HBBRG algorithm with the aim to get better results for remove pectoral muscles. The proposed algorithms were tested on all the Mammographic Image Analysis Society (MIAS) database images, and the results showed a significant advantage in the HBBRG algorithm compared to other algorithms as it achieved results in over 98% to completely remove the pectoral muscles of all types of images.

Pectoral Muscles Removal in Mammogram Image by Hybrid Bounding Box and Region Growing Algorithm

In the majority of gene expression investigations, selecting relevant genes for sample classification is considered a frequent challenge, with researchers attempting to discover the minimum feasible number of genes while yet achieving excellent predictive performance. Various gene selection methods employ univariate (gene-by-gene) gene relevance rankings as well as arbitrary thresholds for selecting the number of genes, are only applicable to 2-class problems and use gene selection ranking criteria unrelated to the algorithm of classification. A modified random forest (MRF) algorithm depending on the meerkat clan algorithm (MCA) is provided in this work. It is one of the swarm intelligence algorithms and one of the most significant machine learning approaches in the decision tree. MCA is used to choose characteristics for the RF algorithm. In information systems, databases, and other applications, feature selection imputation is critical. The proposed algorithm was applied to three different databases, where the experimental results for accuracy and time proved the superiority of the proposed algorithm over the original algorithm. 

Enas Mohammed Hussein Saeed

Papers

Pectoral Muscles Removal in Mammogram Image by Hybrid Bounding Box and Region Growing Algorithm

Classification of mammograms based on features extraction techniques using support vector machine

Hybrid features selection method using random forest and meerkat clan algorithm