Mohammed Ali Roula

Round-Robin sequential forward selection algorithm for prostate cancer classification and diagnosis using multispectral imagery

This paper proposes an automatic classification system for the use in prostate cancer diagnosis. The system aims to detect and classify prostatic tissue textures captured from microscopic samples taken from needle biopsies. Biopsies are usually analyzed by a trained pathologist with different grades of malignancy typically corresponding to different structural patterns as well as apparent textures. In the context of prostate cancer diagnosis, four major groups have to be accurately recognized: stroma, benign prostatic hyperplasia, prostatic intraepithelial neoplasia, and prostatic carcinoma. Recently, multispectral imagery has been proposed as a new image acquisition modality which unlike conventional RGB-based light microscopy allows the acquisition of a large number of spectral bands within the visible spectrum, resulting in a large feature vector size. Many features in the initial feature set are irrelevant to the classification task and are correlated with each other, resulting in an increase in the computational complexity and a reduction in the recognition rate. In this paper, a Round-Robin (RR) sequential forward selection RR-SFS is used to address these problems. RR is a technique for handling multi-class problems with binary classifiers by training one classifier for each pair of classes. The experimental results demonstrate this finding when compared with classical method based on the multiclass SFS and other ensemble methods such as bagging/boosting with decision tree (C4.5) classifier where it is shown that RR-SFS method achieves the best results with a classification accuracy of 99.9%.

Parallelization methods for implementation of a magnetic induction tomography forward model in symmetric multiprocessor systems

This paper describes four parallelization approaches used in a finite-difference-based electromagnetic modeller for application in magnetic induction tomography (MIT) and suitable for implementation on computer systems with symmetric multiprocessor (SMP) architecture. The approaches include: (i) splitting by coils using a distributed memory approach, (ii) splitting by physical domain using a distributed memory approach, (iii) splitting by physical domain using hybrid distributed/shared memory approach and (iv) splitting by both coils and physical domain using multi-level distributed and shared memory approaches respectively. All four approaches were implemented and tested on an IBM SP supercomputer. Coil parallelization was the most efficient method due to low inter-processor communication requirements but was limited by the number of coils in the MIT system. Approaches (ii) and (iii) allowed a larger number of processors to be employed but the efficiency versus number of processors was found to drop at a faster rate in comparison to (i). The fourth approach both allowed a larger number of processors to be employed and was found to provide higher efficiency than the parallelization by physical domain only. This multi-level hybrid approach therefore appears to offer an effective parallelization method for implementation of the MIT forward model on SMP clusters.

Approaches for improving image quality in magnetic induction tomography

Magnetic induction tomography (MIT) is a contactless and non-invasive method for imaging the passive electrical properties of objects. Measuring the weak signal produced by eddy currents within biological soft tissues can be challenging in the presence of noise and the large signals resulting from the direct excitation-detection coil coupling. To detect haemorrhagic stroke in the brain, for instance, high measurement accuracy is required to enable images with enough contrast to differentiate between normal and haemorrhaged brain tissues. The reconstructed images are often very sensitive to inevitable measurement noise from the environment, system instabilities and patient-related artefacts such as movement and sweating. We propose methods for mitigating signal noise and improving image reconstruction. We evaluated and compared the use of a range wavelet transforms for signal denoising. Adaptive regularization methods including L-curve, generalized cross validation (GCV) and noise estimation were also compared. We evaluated all these described methods with measurements of in vitro tissues resembling a peripheral haemorrhagic cerebral stroke created by placing a bio-membrane package filled with 10 ml blood in a swine brain of 100 ml. We show that wavelet packet denoising combined with adaptive regularization can improve the quality of reconstructed images.

Brain computer interfacing using humour and memory recall

Many Brain computer interfaces use active mental tasks such as a users imagined hand movement to generate a signature EEG signal calibrated to a specific command. This is often specific to the individual who has trained the BCI (Brain Computer Interface) over a period of time. To allow multiple users to use an interface without training will help facilitate transferability across subjects, especially with patients whose disability impairs the possibility of full training. The current study examines the use of the memory recall of humour and moving imagery for activating braining computer interfaces, results show that humour is a response that is classifiable for BCI, with high success rates when used with ones own calibration signature (82.9%) or someone elses calibrated signature (80.0%).

Fuzzy thresholding of color images using dempster-shafer theory

This paper is concerned with the development of a novel color image thresholding technique using fuzzy thresholding and Dempster-Shaferpsilas theory based fusion. The color bands of a given image are fuzzified by means of a fuzzy thresholding technique which uses an S-shape membership function and a linear index of fuzziness based fuzzy measure. The resulting fuzzy maps are used to determine the mass functions of the hypotheses representing the classes for each pixel. These hypotheses are then combined using the orthogonal sum rule of the Dempster-Shafer theory to compute the final threshold map. The results show that the proposed algorithm yields a superior performance to its counterpart methods which are based on crisp and fuzzy techniques.

Effects of system parameters on particle statistics in aerosol charge and size measurement in oscillatory electric fields

In this paper, a numerical model of aerosol Particle Charge and Size Analyzer is discussed together with the simulation results. In order to determine the optimal range of system parameters, the effect of drive frequency, strength of the electric field, mean flow velocity on the particle count percentage have been investigated based on the specific criteria in sine and square wave excitation fields. The optimal range of system parameters for both excitation methods has been determined with the recommended values proposed for sine and square wave excitation systems.

Effect of Anodal-tDCS on Event-Related Potentials: A Controlled Study

We aim to measure the postintervention effects of A-tDCS (anodal-tDCS) on brain potentials commonly used in BCI applications, namely, Event-Related Desynchronization (ERD), Event-Related Synchronization (ERS), and P300. Ten subjects were given sham and 1.5 mA A-tDCS for 15 minutes on two separate experiments in a double-blind, randomized order. Postintervention EEG was recorded while subjects were asked to perform a spelling task based on the “oddball paradigm” while P300 power was measured. Additionally, ERD and ERS were measured while subjects performed mental motor imagery tasks. ANOVA results showed that the absolute P300 power exhibited a statistically significant difference between sham and A-tDCS when measured over channel Pz (p = 0.0002). However, the difference in ERD and ERS power was found to be statistically insignificant, in controversion of the the mainstay of the litrature on the subject. The outcomes confirm the possible postintervention effect of tDCS on the P300 response. Heightening P300 response using A-tDCS may help improve the accuracy of P300 spellers for neurologically impaired subjects. Additionally, it may help the development of neurorehabilitation methods targeting the parietal lobe.

Cellular proteomic characterization using Active Shape and Non-Gaussinan stochastic texture models

This paper presents a method for the systematical extraction cellular parameters from imaging proteomic datasets in a way suitable for subsequent biological modeling and simulation. This was achieved by capturing the spatial boundaries of cell structures as well as the distribution of its constituents. The model uses the Active Shape Models to parameterize the shape of cellular structures and the Non-Gaussian Texture Model to parameterize spatial distribution of sub-cellular material. Results show the model can extract then generate faithful representations of cellular shapes and textures for a variety of cell types and protein expressions and hence could offer a natural spatial framework for current research on simulating and predicting sub-cellular processes.

A parallel implementation of a Magnetic Induction Tomography: Image reconstruction algorithm on the ClearSpeed Advance accelerator board

Magnetic Induction Tomography (MIT) is a relatively new contactless imaging modality which aims at reconstructing conductivity and permittivity distributions within objects. One of MITs main challenges is the computational intensity required for image reconstruction in potential industrial and medical applications.