Rosalyn S. Hobson

Fracture detection in traumatic pelvic CT images

Fracture detection in pelvic bones is vital for patient diagnostic decisions and treatment planning in traumatic pelvic injuries. Manual detection of bone fracture from computed tomography (CT) images is very challenging due to low resolution of the images and the complex pelvic structures. Automated fracture detection from segmented bones can significantly help physicians analyze pelvic CT images and detect the severity of injuries in a very short period. This paper presents an automated hierarchical algorithm for bone fracture detection in pelvic CT scans using adaptive windowing, boundary tracing, and wavelet transform while incorporating anatomical information. Fracture detection is performed on the basis of the results of prior pelvic bone segmentation via our registered active shape model (RASM). The results are promising and show that the method is capable of detecting fractures accurately.

An impedimetric biosensor based on PC 12 cells for the monitoring of exogenous agents

The effect of exogenous agents on the complex impedance of PC 12 cells that were cultured to confluency on 250-mum gold microdot electrodes fabricated within 8-well cell culture biochips was studied. Surface attachment of PC 12 cells to gold microelectrodes was accomplished using cysteamine SAMs covalently derivatized with laminin. The impedimetric response of PC 12 cells that undergo calcium exocytosis in the presence of calcimycin, nifedipine, mannitol and carbachol were identified. Treatment with carbachol induces muscarinic receptor-dependent rises in free cytosolic Ca(2+). Experiments with calcimycin and nifedipine were carried out to clarify the relationship between these two receptor-triggered events. In particular, it is believed to mediate intracellularly the release of Ca(2+) from non-mitochondrial stores. We also examined cellular impedance responsiveness of PC 12 cells in response to phenotypic alteration especially with regard to modulation of ion fluxes using nerve growth factor (NGF), dexamethasone and forskolin. Our results demonstrate that a change in electrophysiological behavior, such as exocytosis of cytosolic Ca(2+) is detectable using impedance spectroscopy, and therefore support the results of impedance fluctuation to be attributed to ion-fluxes.

Image processing and machine learning for diagnostic analysis of microcirculation

This study focuses on detection of capillaries and small blood vessels in the videos recorded from the lingual surface using Microscan SDF system. The purpose of this study is to quantitatively monitor and assess the changes that occur in microcirculation during resuscitation period. The results assist physicians in making diagnostically and therapeutically important decisions such as determination of the effectiveness of the resuscitation process. The proposed algorithm applies advanced digital image processing methods to provide quantitative assessment of video signals for detection and characterization of capillaries. The objective of the algorithm is to segment capillaries, estimate the presence and velocity of Red Blood Cells (RBCs), and identify the distribution of blood flow in capillaries for a variety of normal and abnormal cases. The algorithm first, stabilizes each frame to follow the variations in the consecutive frames. Then, time-averaging techniques are applied to the frames to reduce the motion artifact. Histogram equalization, wavelet transform, and median filtering are the subsequent steps applied to accurately detect the blood vessels in each frame. In order to estimate the velocity of RBCs, space time diagrams are obtained through cross-correlation calculations. This study aims to reduce the human interaction as well as the computation time.

Modeling student retention in science and engineering disciplines using neural networks

Attracting more students into science and engineering disciplines concerned many researchers for decades. Literature used traditional statistical methods and qualitative techniques to identify factors that affect student retention up most and predict their persistence. In this paper we developed two neural network models using a feed-forward backpropagation network to predict retention for students in science and engineering fields. The first model is used to predict incoming freshmen retention and identify correlated pre-college factors. The second model is to classify freshmen groups into three classes: at-risk, intermediate, and advanced students. With total of 338 samples used, 70.1% of students classified correctly.

The changing face of classroom instructional methods: service learning and design in a robotics course

Service-learning is a form of instruction which uses community service activities as part of the medium for learning. There are two components: service which actively engages the student in community service; and focused-directed learning. Service learning enhances the engineering curriculum by linking engineering directly to improving society, which makes the profession more appealing and more diverse. At Virginia Commonwealth University (VCU), service-learning has been incorporated into the robotics course. VCU students work with other students from local high schools to design and construct a mobile robot within time, materials and cost constraints. These robots compete in the For Inspiration and Recognition of Science and Technology: FIRST Competition. This paper describes service-learning and the robotics course and how the two are incorporated to enhance the educational experience of the VCU and high school students and provide a service to the Richmond Virginia community.

Isolated speech recognition using artificial neural networks

In this project artificial neural networks are used as research tool to accomplish automated speech recognition of normal speech. A small size vocabulary containing the words YES and NO is chosen. Spectral features using cepstral analysis are extracted per frame and imported to a feed-forward neural network which uses a backpropagation with momentum training algorithm. The network is trained to recognize and classify the incoming words into the respective categories. The output from the neural network is loaded into a pattern search function, which matches the input sequence with a set of target word patterns. The level of variability in input speech patterns limits the vocabulary and affects the reliability of the network. The results from the first stage of this. work are satisfactory and thus the application of artificial neural networks in conjunction with cepstral analysis in isolated word recognition holds promise.

A new hierarchical method for multi-level segmentation of bone in pelvic CT scans

Pelvic bone segmentation is a vital step in analyzing pelvic CT images and assisting physicians with diagnostic decisions in traumatic pelvic injuries. A new hierarchical segmentation algorithm is proposed using a template-based best shape matching method and Registered Active Shape Model (RASM) to automatically extract pelvic bone tissues from multi-level pelvic CT images. A novel hierarchical initialization process for RASM is proposed. 449 CT images across seven patients are used to test and validate the reliability and robustness of the proposed method. The segmentation results show that the proposed method performs better with higher accuracy than standard ASM method.

Artificial Neural Network for Temporal Impedance Recognition of Neurotoxins

The design, development and in-vitro evaluation of an impedimetric neurotoxicity cell-based biosensor that is designed for real time monitoring of changes in electrophysiological behavior under the influence of neurotoxins is described. The electrical cell impedance sensing (ECIS) system [ECIS 8W1E element array of gold electrodes] is used as a substrate for the culture of rat pheochromocytoma (PC 12) cells. The neurotoxicity biosensor is a microfabricated solid state device that mimics the natural environment of PC 12 cells that are responsive to neurotoxins. The PC 12 neurotoxicity biosensors are complemented by artificial neural networks (ANNs) to recognize the impedance profiles of the cells under the influence of a neurotoxin. The neurotoxins were rotenone (Rot), okadaic acid (OA) and peroxynitrite (Per), which are all known to induce cell death in PC 12 cells. Three multilayer feedforward artificial neural network models were developed using a back-propagation algorithm for pattern recognition of neurotoxins. The neurotoxin network (NTN) and the neurotoxin concentration network (NTCN), were trained with data from all the neurotoxins and the cascade network (NTN_NTCN) was developed by combining both the NTN and NTCN. The cascade network was developed to screen against false positives. The neurotoxicity biosensor coupled with these networks allowed for the action of unknown agents (neurotoxins) to be deduced by the measured cellular response. Using back-propagation ANNs to distinguish neurotoxins under the cascade network, the highest success recognition rate for concentration identification were 96% for peroxynitrite, 88% for rotenone, and 96% for okadaic acid. The recognition rate for neurotoxin identification was 98%. The ANN models required less than ten minutes to train and demonstrated that back-propagation ANNs can be handled by commercially-available computers to train and assimilate neurotoxin impedance information, permitting high success rates in the neurotoxin recognition problems.

Fracture detection and quantitative measure of displacement in pelvic CT images

Traumatic pelvic injury is a severe and common injury in the United States. The automatic detection of fractures in pelvic CT images is a significant contribution for assisting physicians in making faster and more accurate patient diagnostic decisions and treatment planning. However, due to the low resolution and quality of the original images, the complexity of pelvic structures, and the difference in visual characteristics of fracture by their location, it is difficult to detect and accurately locate the pelvic fractures and determine the severity of the injury. In this paper, an automatic hierarchical algorithm for detecting pelvic bone fractures in CT scans is proposed. The algorithm utilizes symmetric comparison, adaptive windowing, boundary tracing, wavelet transform. Also, the quantitative measure of fracture severity in pelvic CT scans is defined. The results are promising, demonstrating that the proposed method is capable of automatically detecting both major and minor fractures accurately, shows potential for clinical application. Statistical results also indicate the superiority of the proposed method.

An automated method for hemorrhage detection in traumatic pelvic injuries

Hemorrhage is the main cause of deaths that occurs within first 24 hours after a traumatic pelvic injury. Therefore, it is very important to determine hemorrhage quickly. Hemorrhages are detected using a CT scan. However, it is very time consuming for physicians to look for hemorrhage in all CT slices. Therefore, an automated system is needed. This paper proposes an automated hemorrhage detection technique by incorporating anatomical information of pelvic region. The results showed method performs comparably to manual methods. A statistical test is conducted to see if the volume of hemorrhage detected using this technique is significantly different from the volume assessed manually.