Mehmed Ozkan

Neural-network-based segmentation of multi-modal medical images: a comparative and prospective study

This work presents an investigation of the potential of artificial neural networks for classification of registered magnetic resonance and X-ray computer tomography images of the human brain. First, topological and learning parameters are established experimentally. Second, the learning and generalization properties of the neural networks are compared to those of a classical maximum likelihood classifier and the superiority of the neural network approach is demonstrated when small training sets are utilized. Third, the generalization properties of the neural networks are utilized to develop an adaptive learning scheme able to overcome interslice intensity variations typical of MR images. This approach permits the segmentation of image volumes based on training sets selected on a single slice. Finally, the segmentation results obtained both with the artificial neural network and the maximum likelihood classifiers are compared to contours drawn manually.

GSM Base Station Electromagnetic Radiation and Oxidative Stress in Rats

The ever increasing use of cellular phones and the increasing number of associated base stations are becoming a widespread source of nonionizing electromagnetic radiation. Some biological effects are likely to occur even at low-level EM fields. In this study, a gigahertz transverse electromagnetic (GTEM) cell was used as an exposure environment for plane wave conditions of far-field free space EM field propagation at the GSM base transceiver station (BTS) frequency of 945 MHz, and effects on oxidative stress in rats were investigated. When EM fields at a power density of 3.67 W/m2 (specific absorption rate = 11.3 mW/kg), which is well below current exposure limits, were applied, MDA (malondialdehyde) level was found to increase and GSH (reduced glutathione) concentration was found to decrease significantly (p < 0.0001). Additionally, there was a less significant (p = 0.0190) increase in SOD (superoxide dismutase) activity under EM exposure.

Prediction of bank failures in emerging financial markets: an ANN approach

Purpose - The recent financial crises in the world have brought attention to the need for a new international financial architecture which rests on crisis prevention, crisis prediction and crisis management. It is therefore both desirable and vital to explore new predictive techniques for providing early warnings to regulatory agencies. The purpose of this study is to propose a new technique to prevent future crises, with reference to the last banking crises in Turkey. Design/methodology/approach - ANN is utilized as an inductive algorithm in discovering predictive knowledge structures in financial data and used to explain previous bank failures in the Turkish banking sector as a special case of EFMs (emerging financial markets). Findings - The empirical results indicate that ANN is proved to differentiate patterns or trends in financial data. Most of the bank failures could be predicted long before, with the utilization of an ANN classification approach, but more importantly it could be proposed to detect early warning signals of potential failures, as in the case of the Turkish banking sector. Practical implications - The regulatory agencies could use ANN as an alternative method to predict and prevent future systemic banking crises in order to minimize the cost to the economy. Originality/value - This paper reveals that the ANN approach can be proposed as a promising method of evaluating financial conditions in terms of predictive accuracy, adaptability and robustness, and as an alternative early warning method that can be used along with the most common alternatives such as CAMEL, financial ratio and peer group analysis, comprehensive bank risk assessment, and econometric models.

Defining a neural network controller structure for a rubbertuator robot

Rubbertuator (Rubber-Actuator) robot arm is a pneumatic robot, unique with its lightweight, high power, compliant and spark free nature. Compressibility of air in the actuator tubes and the elastic nature of the rubber, however, are the two major sources of increased non-linearity and complexity in motion control. Soft computing, exploiting the tolerance of uncertainty and vagueness in cognitive reasoning has been offering easy to handle, robust, and low-priced solutions to several non-linear industrial applications. Nonetheless, the black-box approach in these systems results in application specific architectures with some important design parameters left for fine tuning (i.e. number of nodes in a neural network). In this study we propose a more systematic method in defining the structure of a soft computing technique, namely the backpropagation neural network, when used as a controller for rubbertuator robot systems. The structure of the neural network is based on the physical model of the robot, while the neural network itself is trained to learn the trajectory independent parameters of the model that are essential for defining the robot dynamics. The proposed system performance was compared with a well-tuned PID controller and shown to be more accurate in trajectory control for rubbertuator robots.

Assessment of surgeon's stress level and alertness using EEG during laparoscopic simple nephrectomy

Laparoscopic simple nephrectomy (LSN) is an accepted treatment modality for nonfunctioning kidneys. Besides decreased postoperative morbidity, LSN is an advantage with decreased analgesic requirements and convalescence. LSN is a highly stressful operation, and the procedure requires high concentration level and experience. Emotions recognized from Electroencephalogram (EEG) may lead to detect the real emotions of the human. In this study, we proposed a subject-dependent stress level detection from EEG using the (Fpz beta/alpha) ratio to recognize high and low dominance levels of feelings based on the 2D Valence-Arousal model. The stress level of the surgeon is monitored via EEG during the operation. The most stressful phase of LSN and its change over time are determined using wireless EEG headset with real-time measurements. The aim here is to monitor and utilize objective information on the mental effort and stress demanded.

Classification of ECG Arrythmia beats with Artificial Neural Networks

In this study, Electrocardiographic(ECG) Arrythmias were classified by using Artificial Neural Networks (ANN). During the training process of ANN, the ECG recordings from MIT BIH Arrythmia database are used as a reference. 24 recordings out of 48 30 minutes recordings in this database were used for data extraction. In order to have more realistic data, the extractons were made from different recordings, and, the typical ECG signals with acceptable amount of noise were included. The arrhythmia samples that are extracted from the database were prepreprocessed to create input sets to train ANNs. The Fourier Transforms of a predefined window of signals were taken as a feature extraction method. As a result of this study, 5 types of ECG signals (Ventricular Tachicardy, Left Bundle Branch Block, Right Bundle Branch Block, Atrial Fibrillation, Normal ECG) were labeled with 82% accuracy.

Multispectral magnetic resonance image segmentation using neural networks

The design, implementation, and preliminary testing of a computer system for automatic multispectral magnetic resonance imaging analysis is presented. The modular structure of the system permits easy comparison between various classification algorithms. The classification accuracy of traditional statistical pattern-recognition algorithms is compared to the results that can be obtained with neural networks of different topologies. Quantitative (confusion matrices) as well as visual (segmented images) results of a study performed on sets of normal and pathological images are presented. Images segmented with a neural network classifier (NNC) appear less noisy than images segmented with a maximum likelihood classifier (MLC), and it has been observed that the NNC is less sensitive to the selection of the training sets than the MLC

The development of a new orthosis (neuro-orthosis) for patients with carpal tunnel syndrome: Its effect on the function and strength of the hand

Static wrist orthoses (SWOs) are used in the treatment of carpal tunnel syndrome (CTS) with some drawbacks. As an alternate approach to SWOs, an active closed-loop wrist control strategy based on the principles of functional electrical stimulation was proposed to limit wrist movements. The purpose of the study was to determine whether the proposed ‘neuro-orthosis’ (NeO) system resulted in less restriction in the hand compared to clinically accepted custom-made SWOs while limiting the wrist movements. A case-control study was designed to determine the specific effects of the system on patients with CTS. A total of 24 right-handed female volunteers (12: CTS, 12: healthy) participated in the study. Function, dexterity, and strengths were measured under three different testing conditions: without orthosis, with SWO, and with the NeO system. Maximum angles in one subtest while the NeO system was on and off and general discomfort levels in SWO and NeO test conditions were recorded. The NeO system resulted in less restriction with respect to SWO and provided considerable angular limitation compared to placebo. It was concluded that the proposed prototype control system can be a good candidate to limit the wrist movements in place of SWOs with a better degree of freedom in patients with CTS.

24 DOF EMG controlled hybrid actuated prosthetic hand

A complete mechanical design concept of an electromyogram (EMG) controlled hybrid prosthetic hand, with 24 degree of freedom (DOF) anthropomorphic structure is presented. Brushless DC motors along with Shape Memory Alloy (SMA) actuators are used to achieve dexterous functionality. An 8 channel EMG is used for detecting 7 basic hand gestures for control purposes. The prosthetic hand will be integrated with the Neural Network (NNE) based controller in the next phase of the study.

Interaction of Ligament Bundles and Articular Contacts for the Simulation of Passive Knee Flexion

The purpose of this study is to investigate the effects of anterior bundle of ACL (aACL), anterior portion of PCL (aPCL), anterior and deep portions of MCL (aMCL, dMCL) and the tibio-femoral articular contacts on to passive knee motion. A three-dimensional simplistic anatomical dynamic model, based on the literature was used as a reference. This reference model attaches the bundles of the ligaments on medial and the lateral spherical condyles of the femur and tibial plateau giving us a representation close enough to a normal natural tibio-femoral joint, but does not allow to study abnormalities of the knee kinematics due to the assumptions of the femur shape. The proposed three-dimensional dynamic tibio-femoral model, however includes the isometric fascicles, aACL, aPCL, aMCL, dMCL, and irregularly shaped medial- lateral contact surfaces. The approach taken in this model is capable of ligament and bone surface modifications that will enable us to analyze bone shape and ligament related abnormalities of knee kinematics.