Siti Noraini Sulaiman

Adaptive fuzzy-K-means clustering algorithm for image segmentation

Clustering algorithms have successfully been applied as a digital image segmentation technique in various fields and applications. However, those clustering algorithms are only applicable for specific images such as medical images, microscopic images etc. In this paper, we present a new clustering algorithm called Adaptive Fuzzy-K-means (AFKM) clustering for image segmentation which could be applied on general images and/or specific images (i.e., medical and microscopic images), captured using different consumer electronic products namely, for example, the common digital cameras and CCD cameras. The algorithm employs the concepts of fuzziness and belongingness to provide a better and more adaptive clustering process as compared to several conventional clustering algorithms. Both qualitative and quantitative analyses favour the proposed AFKM algorithm in terms of providing a better segmentation performance for various types of images and various number of segmented regions. Based on the results obtained, the proposed algorithm gives better visual quality as compared to several other clustering methods.

Denoising-based clustering algorithms for segmentation of low level salt-and-pepper noise-corrupted images

Clustering algorithm is a widely used segmentation method in image processing applications. The algorithm can be easily implemented; however in the occurrence of noise during image acquisition, this might affect the processing results. In order to overcome this drawback, this paper presents a new clustering-based segmentation technique that may be able to find different applications in image segmentation. The proposed algorithm called Denoising-based (DB) clustering algorithm has three variations namely, Denoising-based-K-means (DB-KM), Denoising-based-Fuzzy C-means (DB-FCM), and Denoising-based-Moving K-means (DB-MKM). The proposed DB-clustering algorithms are able to minimize the effects of the Salt-and-Pepper noise during the segmentation process without degrading the fine details of the images. These methods incorporate a noise detection stage to the clustering algorithm, producing an adaptive segmentation technique specifically for segmenting the noisy images. The results obtained quantitatively and qualitatively have favored the proposed DB-clustering algorithms, which consistently outperform the conventional clustering algorithms in segmenting the noisy images. Thus, these DB-clustering algorithms could be possibly used as pre- or post-processing (i.e., segmenting images into regions of interest) in consumer electronic products such as television and monitor with their capability of reducing noise effect.

Capability of new features from FTIR spectral of cervical cells for cervical precancerous diagnostic system using MLP networks

The applicability and reliability of Infrared (IR) spectroscopy to distinguish normal and abnormal cells has opened this research to obtain new features from IR spectral of cervical cells to be fed into multilayered perceptrons (MLP) networks. In order for neural networks to be used as cervical precancerous diagnostic system, the features of cervical cell were used as inputs for neural networks and the classification of cervical cell types were used as output target. For cervical cell classification, this study proposes new features of cervical cell spectrum that are suitable and can be used as inputs for neural networks. The new cervical cell features were extracted from ThinPrep® spectrum and their applicability were tested by using seven types of MLP training algorithm. The MLP network trained using Levenberg-Marquardt Backpropogation (trainlm) algorithm presented the highest accuracy with percentage of 97.3%. The result shows that the proposed features i.e. area under spectrum at 1800-1500 cm−1, area under spectrum at 1200–1000 cm−1, area under spectrum at 1800-950 cm−1, height of slope at 1650-1550 cm−1, corrected area of protein band at 1590-1490 cm−1, corrected area of glycogen band at 1134-985 cm−1, corrected peak height protein (H1545) and corrected peak height glycogen (H1045) are applicable to be fed as input to neural network for cervical spectra classification in cervical precancerous diagnostic system.

Improvement of Features Extraction Process and Classification of Cervical Cancer for the NeuralPap System

Abstract Cervical cancer has caused many deaths each year. Screening tests, such as Pap smear test used for the detection of the precancerous stage are able to avoid the occurrence of cervical cancer. However, the Pap smear test has several disadvantages such as less effective slides preparation and human error. Therefore, a computer-aided diagnosis system is introduced as a solution to the problem. One of the diagnostic systems that has been built is NeuralPap. However, the NeuralPap performance is limited by several constraints. This research proposed several new image processing algorithms to reduce these constraints. The Adaptive Fuzzy-k-Means (AFKM) clustering algorithm is proposed to replace the Moving k-Means (MKM) to segment Pap smear images into the nucleus, cytoplasm and background regions. Next, the feature extraction algorithm based on pseudo colouring called the Pseudo Colour Feature Extraction (PCFE) manual and Semi-Automatic PCFE are designed to replace the Region Growing Based Feature Extraction (RGBFE) which uses monochromatic images. This research is a step forward compared with the NeuralPap system by proposing the feature extraction algorithm for overlapping cells by combining the concept of colour space with Semi-Automatic PCFE algorithm. In addition, this research has also suggested the AFKM algorithm as a new centre positioning algorithm for the Radial Basis Function (RBF) and Hybrid RBF (HRBF) networks replacing the MKM algorithm. The entire proposed algorithm has been proven to produce better performance than the corresponding algorithm used in the NeuralPap. In addition, the combination of all algorithms has managed to increase the accuracy of the classification of cervical cancer to 76.35%, compared with 73.40% which is obtained from the previous NeuralPap system.

Overlapping cells separation method for cervical cell images

Image analysis is one of the common application fields in medical especially in cytology. Visual interpretation is the core for most medical diagnostic procedure. This process is tedious especially in the existence of overlapping cells therefore it is crucial to split the overlapping cells into single ones. This study proposes an overlapping cells separation method for separating the overlapping cells. The proposed technique integrates Edge Detection process and Pseudo-Color technique, with Color Space Extraction employed at preprocessing stage. First, the color space concept is applied to extract the original image of Pap smear into red plane, green plane and blue plane. Then the Seed Based Region Growing technique is applied to find boundaries of the cells. Pseudo-color technique is then embedded to the demarcated region to determine each part of the cell; nucleus, cytoplasm and background. The proposed technique is capable to distinguish each cervical cell from overlapping cervical cells image. Therefore, the resultant image will be more useful for further analysis.

Pseudo Color Features Extraction technique for cervical cancer of Pap smear images

Cervix cancer is the most common gynecological malignancy and second most common cancer among female in Malaysia after breast cancer. The objective of this study is to extract the size of nucleus and cytoplasm, as well as gray level values of cervical cells from ThinPrep images so that accurate value of those parameters can easily be obtained. An alternative approach of extracting features for Pap smear cytology images i.e., by using Seed Based Region Growing technique and Pseudo Coloring is proposed in this study. The technique is called Pseudo Color Feature Extraction (PCFE). A correlation test is applied between data extracted using the proposed algorithm and data extracted manually by cytotechnologists. The technique operates well on cervical cells images with correlation values approaching 1.0 which indicates a strong positive correlation. The strongest relationship is the size of cytoplasm with correlation factor of 0.988595 and the next strongest relationship is its gray level with correlation factor of 0.981534. Such results indicate that the proposed technique is suitable and has high capability to be used as an image extraction technique for extracting cervical cells features as well as acts as a filter and a segmentation tool. This would assist cytopathologists and cytotechnologists in the cervical cancer screening process by providing accurate value of size and gray level of nucleus and cytoplasmic features.

De-noising of noisy MRI brain image using the switching-based clustering algorithm

Magnetic Resonance Image is one of the technologies used for diagnosing brain cancer. Radiographers use the information obtained from MRI images to diagnose the disease and plan further treatment for the patient. MRI images are always corrupted with noise. Removing noise from images is crucial but it is not an easy task. Filtering algorithm is the most common method used to remove noise. A segmentation technique is normally used to process the image in order to detect the abnormality that has been observed, specifically in the brain. However, segmentation alone would be best to implement when the images are in good condition. In the case where the images are corrupted with noise, there are pre-processing steps that need to be implemented first before we can proceed to the next task. Therefore, in this project, we have proposed a simpler method that can de-noise and at the same time segment the image into several significant regions. The proposed method is called the switching-based clustering algorithm. The algorithm is implemented on the MRI brain images which are corrupted with a certain level of salt-and-pepper noise. During the segmentation process, the results show that the proposed algorithm has the ability to minimize the effect of noise without degrading the original images. The density of noise in the MRI images varies from 5% to 20%. The results are compared with the conventional clustering algorithm. Based on the experimental result obtained, the switching-based algorithm provides a better segmentation performance with fewer noise effects than the conventional clustering algorithm. Quantitative and qualitative analyses have shown positive results for the proposed switching-based clustering algorithm.

Modified hybrid median filter for removal of low density random-valued impulse noise in images

Random-valued impulse noise (RVIN) is a randomly distributed noise of two significant pixel values that degrades the quality of digital images during acquisition, processing, and storage. It is a variation of the salt-and-pepper or fixed-valued impulse noise (FIN) which instead of the black and white pixel, the noisy pixel value can be anywhere in the range of the grey level pixel. This paper introduces a new filter which is a modified hybrid median filter for removal of RVIN from digital images. The proposed filter is based on similar standard median filters and an improvement upon the hybrid median filter which make use of neighborhood pixels in reducing RVIN from the image. This filter operates using a window size of 3×3 and uses values in the window with a modified hybrid median algorithm to replace the targeted pixel during the filtering process. This technique has proven to be capable of overcoming the shortcomings of standard median filter as well as improve upon the hybrid median filter in restoring image details and in operating at higher noise density.

Performance of Hybrid Radial Basis Function network: Adaptive Fuzzy K-Means versus Moving k-Means Clustering as centre positioning algorithms on cervical cell precancerous stage classification

Cervical cancer been known to be the cause of many deaths each year. Screening tests, such as the Pap smear test used for the detection of the precancerous stage are able to avoid the occurrence of cervical cancer. However, the Pap smear test does have some disappointing disadvantages such as the fact that it has less effective slide preparation and also that it is laden with human error. Therefore, a computer-aided diagnosis system is introduced as a solution to the problem. Recently, artificial neural networks have been widely implemented as a cervical cancer diagnosis system i.e. to classify cervical cancer into normal and abnormal cells. In this recent study, neural network architecture i.e. the Hybrid Radial Basis Function (HRBF) network with Adaptive Fuzzy K-Means Clustering (AFKM) asa centre positioning algorithm is used to diagnose cervical cancer. Four extracted features of cervical cell are used as input data to the networks, which are the size of nucleus and cytoplasm and its grey level. Cells from normal, Low-grade Squamous Intraepithelial Lesion (LSIL) and High-grade Squamous Intraepithelial Lesion (HSIL) categories are used as the training as well as the testing data. The data are fed randomly into the neural networks via 5-folds cross validation techniques. The network performance is compared with the HRBF network with the Moving K-Means algorithm as the centre positioning algorithm. The proposed network produces better accuracy, sensitivity and specificity which illustrates the promising capability of the network to be implemented as cervical cancer diagnosis system for Pap test performance improvement.

WMH Detection Using Improved AIR-AHE-Based Algorithm for Two-Tier Segmentation Technique

The segmentation of magnetic resonance imaging (MRI) brain images could be implemented using any technique, either automatic or manual. The different methods commonly show different results because their performance relies on the segmentation precision and accuracy. In this paper, a new image segmentation algorithm is proposed based on k-means and AIR-AHE clustering algorithm to automatically segment and classify WMH severity in brain white matter region. The objective of this new segmentation algorithm is to minimize the false positive (FP) in white matter hyper-intensity (WMH) detection and hence will increase the WMH detection accuracy in MRI images. The proposed algorithm is implemented on two-tier segmentation system by identifying the edge of WMH and WM boundary for image mapping purpose. T2-weighed imaging (T2-WI) and fluid-attenuated inversion recovery (FLAIR) MRI sequences are used for mapping most precise WMH region of interest (ROI). From the experimental results, the proposed algorithm produces significant improvement in terms of correct WMH localization and reduces the false WMH detection. Based on the accuracy and capabilities of the proposed algorithm, this algorithm is suitable to be implemented to aid radiologist in the image analysing.