Dissertations / Theses on the topic 'Human Activity Prediction'

Being able to predict and recognize human activities is an essential element for us to effectively communicate with other humans during our day to day activities. A system that is able to do this has a number of appealing applications, from assistive robotics to health care and preventative medicine. Previous work in supervised video-based human activity prediction and detection fails to capture the richness of spatiotemporal data that these activities generate. Convolutional Long short-term memory (Convolutional LSTM) networks are a useful tool in analyzing this type of data, showing good results in many other areas. This thesis’ focus is on utilizing RGB-D Data to improve human activity prediction and recognition. A modified Convolutional LSTM network is introduced to do so. Experiments are performed on the network and are compared to other models in-use as well as the current state-of-the-art system. We show that our proposed model for human activity prediction and recognition outperforms the current state-of-the-art models in the CAD-120 dataset without giving bounding frames or ground-truths about objects.

When moving into a more connected world together with machines, a mutual understanding will be very important. With the increased availability in wear- able sensors, a better understanding of human needs is suggested. The Dart- mouth Research study at the Psychiatric Research Center has examined the viability of detecting and further on predicting human behaviour and complex tasks. The field of smoking detection was challenged by using the Q-sensor by Affectiva as a prototype. Further more, this study implemented a framework for future research on the basis for developing a low cost, connected, device with Thayer Engineering School at Dartmouth College. With 3 days of data from 10 subjects smoking sessions was detected with just under 90% accuracy using the Conditional Random Field algorithm. However, predicting smoking with Electrodermal Momentary Assessment (EMA) remains an unanswered ques- tion. Hopefully a tool has been provided as a platform for better understanding of habits and behaviour.

This research aims to utilise ligand-based target prediction to (i) understand the mechanism of action of African natural products (ANPs), (ii) help identify patterns of phylogenetic use in African traditional medicine and (iii) elucidate the mechanism of action of phenotypically active small molecules and natural products with anti-trypanosomal activity. In Chapter 2 the objective was to utilise ligand-based target prediction to understand the mechanism of action of natural products (NPs) from African medicinal plants used against cancer. The Random Forest classifier used in this work compares the similarity of the input compounds from the natural product dataset with compound-target combinations in the training set. The more similar they are in structure, the more likely they are to modulate the same target. Natural products from plants used against cancer in Africa were predicted to modulate targets and pathways directly associated with the disease, thus understanding their mechanism of action e.g. "flap endonuclease 1" and "Mcl-1". The "Keap1-Nrf2 Pathway" and "apoptosis modulation by HSP70", two pathways previously linked to cancer (which are not currently targeted by marketed drugs, but have been of increasing interest in recent years) were predicted to be modulated by ANPs. In Chapter 3, we aimed to identify phylogenetic patterns in medicinal plant use and the role this plays in predicting medicinal activity. We combined chemical, predicted target and phylogenetic information of the natural products to identify patterns of use for plant families containing plant species used against cancer in African, Malay and Indian (Ayurveda) traditional medicine. Plant families that are close phylogenetically were found to produce similar natural products that act on similar targets regardless of their origin. Additionally, phylogenetic patterns were identified for African traditional plant families with medicinal species used against cancer, malaria and human African trypanosomiasis (HAT). We identified plant families that have more medicinal species than would statistically be expected by chance and rationalised this by linking their activity to their unique phyto-chemistry e.g. the napthyl-isoquinoline alkaloids, uniquely produced by Acistrocladaceae and Dioncophyllaceae, are responsible for anti-malarial and anti-trypanosome activity. In Chapter 4, information from target prediction and experimentally validated targets was combined with orthologue data to predict targets of phenotypically active small molecules and natural products screened against Trypanosoma brucei. The predicted targets were prioritised based on their essentiality for the survival of the T. brucei parasite. We predicted orthologues of targets that are essential for the survival of the trypanosome e.g. glycogen synthase kinase 3 (GSK3) and rhodesain. We also identified the biological processes predicted to be perturbed by the compounds e.g. "glycolysis", "cell cycle", "regulation of symbiosis, encompassing mutualism through parasitism" and "modulation of development of symbiont involved in interaction with host". In conclusion, in silico target prediction can be used to predict protein targets of natural products to understand their molecular mechanism of action. Phylogenetic information and phytochemical information of medicinal plants can be integrated to identify plant families with more medicinal species than would be expected by chance.

De l'industrie aux services, les systèmes intelligents sont amenés à observer, interagir ou encore coopérer avec l'humain. Cette thèse s'inscrit ainsi dans le contexte des méthodes de perception intelligente pour l'analyse de l'humain en utilisant la pose et l'activité qui lui sont associées. En raison de la nature variable et changeante de l'humain, il est difficile d'obtenir une représentation précise des processus guidant ses mouvements et ses actions. Ces difficultés sont accrues lorsqu'il s'agit d'estimer ou de prédire les mouvements ou les activités. Afin de considérer l'incertitude inhérente à l'humain, nous proposons une approche bayésienne pour la perception et l'analyse de l'activité humaine. La première contribution est consacrée à l'estimation simultanée de la pose et de la posture humaine. À l'aide d'une caméra monoculaire et de capteurs portés, nous cherchons à estimer la pose 3D humaine en temps réel. Pour une estimation robuste, une approche de fusion multimodale est proposée en incorporant les mesures de capteurs inertiels portés aux observations caméra. De cette manière, nous outrepassons les ambiguïtés de mesure liées à la caméra et à la dérive inertielle due aux centrales inertielles. Afin de tenir compte de la nature non déterministe du mouvement humain et du caractère non gaussien de la posture, nous choisissons d'utiliser un filtre particulaire. Dans le but de réduire le coût du calcul, nous proposons une architecture composée de deux filtres consécutifs. Un premier filtre estime la posture de manière factorisée uniquement à partir des observations inertielles. Puis un second filtre estime la pose complète à partir de la caméra, en incorporant l'estimation du premier filtre. Notre approche réalise la fusion de manière originale par la construction de la distribution d'échantillonnage du second filtre. Cette architecture permet ainsi d'estimer la pose et la posture de manière simultanée avec un coût calculatoire réduit tout en étant robuste aux occultations et dérives. La seconde contribution s'intéresse à la prédiction de l'activité humaine. Les modèles de Markov cachés se sont montrés efficaces pour l'analyse de l'activité humaine à travers des tâches de segmentation et de reconnaissance de l'activité. Cependant, ils présentent des limites en termes de modélisation les rendant insuffisants pour la prédiction. Nous proposons donc l'utilisation de modèles semi-markoviens pour la prédiction. Ces modèles étendent la définition des modèles de Markov en modélisant de manière explicite la durée passée dans chaque état. Cette modélisation explicite de la durée permet une meilleure modélisation des processus non stationnaires et améliore la capacité prédictive de ces modèles. Notre étude démontre ainsi l'utilité de tels modèles pour la prédiction d'activité avec prise en compte de l'incertitude
From industry to services, intelligent systems are required to observe, interact with, or cooperate with humans. This thesis is therefore set in the context of intelligent perception methods for the analysis of humans, using the pose and activity associated with them. Due to the variable and changing nature of humans, it is difficult to obtain an accurate representation of theprocesses guiding their movements and actions. These difficulties are compounded when it comes to estimating or predicting movements or activities. In order to take account of the uncertainty inherent in humans, we propose a Bayesian approach to the perception and analysis of human activity. The first contribution is dedicated to the simultaneous estimation of human pose and posture. Using a monocular camera and wearable sensors, we aim to estimate human 3D pose in real time. For robust estimation, a multimodal fusion approach is suggested, incorporating measurements from wearable inertial sensors with camera observations. In this way, we overcome measurement ambiguities related to the camera and inertial drift due to inertial units. We use a particle filter so as to take into account the non-deterministic nature of human motion and thenon-Gaussian nature of posture. In order to reduce the computational cost, we put forward an architecture composed of two consecutive filters. A first filter estimates the posture in a factorized way from inertial observations only. Then a second filter estimates the complete pose from the camera, incorporating the estimation of the first filter. Our approach achieves fusion by constructing the sampling distribution of the second filter. This architecture makes it possible to estimate pose and posture simultaneously, at low computational cost, and is robust to cloaking and drift. The second contribution pertains to the prediction of human activity. Hidden Markov models have proved effective for the analysis of human activity through segmentation and activity recognition tasks. However, they have modeling limitations that make them insufficient for prediction. We therefore propose the use of semi-Markovian models for prediction. These models extend the definition of Markov models by explicitly modeling the duration spent in each state. This explicit modeling of duration enables better modeling of non-stationary processes and improves the predictive capability of these models. Our study thus demonstrates the usefulness of such models for activity prediction while taking uncertainty into account

Epilepsy is a neurological disorder affecting 50 million people worldwide. It consists of a large number of syndromes, all of which are characterized by a predisposition to recurrent, unprovoked seizures, while differing by degree of focality, clinical manifestation and many other factors. Despite the prevalence of this disorder, relatively little is known about the basic physiological mechanisms that underlie the seizures themselves. Additionally, roughly 25% of patients are refractory to existing therapies. The need for more highly targeted therapies for focal epilepsies has driven decades of research on seizure prediction. While most of these studies have relied on scalp or intracranial EEG, more recent studies have taken advantage of electrodes that capture single- or multi-unit activity. We utilized a linear microelectrode array to capture multi-unit activity in humans with refractory epilepsy with the expectation that such microscale activity may provide a signal in advance of changes on electroencephalography. Twelve patients underwent long-term monitoring with both clinical electrocorticography (ECoG) and the laminar microelectrode array, which consists of linearly arranged contacts that sample all layers of the human cortex. Multi-unit (300-5000 Hz) power was compared between thirty-minute preictal and interictal time windows. Several parameters characterizing the multi-unit power were compared between preictal and interictal time windows. Parameters included proximity to seizure focus, depth of recording, and directionality of changes in multi-unit power. Optimization of these parameters resulted in a best-performing classifier with sensitivity and specificity of 0.70 and 0.80, respectively. These results demonstrate reproducible increases and decreases in multi-unit activity prior to seizure onset and suggest that multi-unit information may be useful in the development of future seizure prediction systems.

Understanding the control strategies that underlie multijoint limb movements is important to researchers in motor control, robotics, and medicine. Due to dynamic interactions between limb segments, choosing appropriate muscle activations for initiating multijoint arm movements is a complex problem, and the rules by which the nervous system makes such choices are not yet understood. The aim of the dissertation studies was to evaluate some proposed initiation rules based on their ability to correctly predict which shoulder and elbow muscles initiated planar, two-joint arm movements in various directions. Kinematic and electromyographic data were collected from thirteen subjects during pointing movements involving shoulder and elbow rotations in the horizontal plane. One of the rules tested, which is based on statics, predicted that the initial muscle activity at each joint is chosen such that the hand exerts an initial force in the direction of the target, while another rule, based on dynamics, predicted initial muscle activity such that the initial acceleration of the hand is directed toward the target. For both rules, the data contradict the predicted initial shoulder muscle activity for certain movement directions. Moreover, the effects of added inertial loads predicted by the latter rule were not observed when a 1.8 kg mass was added to the limb. The results indicated, however, that empirically derived rules, based on ψ, the target direction relative to the distal segment, could predict which muscles would be chosen to initiate movement in a given direction. Furthermore, the relative timing and magnitude of initial muscle activity at the shoulder and elbow varied systematically with ψ. Thus, the target direction relative to the forearm may be an important variable in determining initial muscle activations for multijoint arm movements. These findings suggest a control scheme for movement initiation in which simple rules suffice to launch the hand in the approximate direction of the target by first specifying a basic motor output pattern, then modulating the relative timing and magnitude of that pattern.

Le principal effet cytotoxique du 5-fluorouracile (5fu) s'exerce par inhibition de la thymidylate synthetase (ts). La formation d'un complexe ternaire intracellulaire entre la ts, un anabolite du 5fu le fluorodeoxyuridine monophosphate (fdump) et un folate reduit, le 5,10-methylenetetrahydrofolate (ch2fh4), bloque la synthese de thymidine et donc la formation d'adn. L'acide folinique (af) potentialise l'effet du 5fu en augmentant le pool intracellulaire de ch2fh4. Une concentration optimale de ch2fh4 sous forme polyglutamatee via la folylpolyglutamate synthetase (fpgs) est necessaire pour une inhibition maximale de la ts. Le 5fu est catabolise par la dihydropyrimidine deshydrogenase (dpd), qui diminue la concentration intratumorale de fdump. Les objectifs de cette etude etaient de tester sur des lignees cellulaires tumorales et des biopsies tumorales de patients, la valeur predictive des activites ts, dpd, fpgs et du ch2fh4 vis a vis de la sensibilite au 5fu et a l'af. Dans les lignees cellulaires, la fpgs est le seul facteur predictif de la sensibilite au 5fu seul ou en presence d'af. L'effet potentialisateur de l'af sur le 5fu est d'autant plus important que le taux de ch2fh4 de base et l'activite fpgs basale sont eleves. Le ch2fh4 intratumoral n'est pas le facteur limitant de l'effet potentialisateur. Dans les tumeurs orl, les patients repondeurs au 5fu ont un taux de ch2fh4 plus eleve et une activite dpd normalisee (dpdtumorale/dpdtissu sain) plus faible que les patients resistants. Au dessus de 1,6 pmole/mg de proteine de ch2fh4, tous les patients sont repondeurs au traitement. Au dessous de 1,6 pmole/min/mg de proteine 52% des patients sont resistants au traitement. Dans le cas des metastases hepatiques de cancers colorectaux, les patients resistants au 5fu ont une activite fpgs plus faible que les repondeurs. 96% des metastases hepatiques dont l'activite fpgs < 1,1 pmole/min/mg de proteine ou dont l'activite ts > 0,32 pmole/min/mg de proteine, sont resistantes a une chimiotherapie a base de 5fu - af

Reconhecimento de atividade humana é uma área de investigação multidisciplinar que tem atraído o interesse de investigadores especializados em aprendizagem automática, visão por computador e medicina. Esta área tem diversas aplicações: sistemas de vigilância, interação homem-máquina, análise de desportos, robôs colaborativos, saúde e automóveis autónomos. Capturar atividade humana apresenta dificuldades técnicas como oclusão, iluminação insuficiente, seguimento erróneo e questões éticas. O movimento humano pode ser ambíguo e com múltiplas intenções. A forma como interagimos com outros seres humanos e objetos cria uma combinação quase infinita de variações de como fazemos as coisas. O objetivo desta dissertação é desenvolver um sistema capaz de reconhecer e prever a atividade humana usando técnicas de aprendizagem automática para extrair significado de características calculadas a partir de articulações do corpo humano capturado pela câmara Kinect. Propomos uma arquitetura hierárquica e modular que realiza segmentação temporal de sequências de ações, anotação semi-supervisionada de sub-atividades utilizando técnicas de clustering, reconhecimento de sub-atividade frame-a-frame em tempo real usando classificadores binários de random decision forests logo a partir dos primeiros instantes da ação e previsão de atividade em tempo real baseada em conditional random fields para modelar a estrutura das sequências de ações para obter as futuras possibilidades. Gravámos um novo conjunto de dados contendo sequências de ações agressivas com um total de 72 sequências, 360 amostras de 8 ações distintas realizadas por 12 sujeitos. Efetuamos testes extensivos com dois conjuntos de dados, comparando o desempenho de reconhecimento de vários classificadores supervisionados treinados com dados anotados manualmente ou com dados anotados de forma semi-supervisionada. Aprendemos como a qualidade dos conjuntos de treino afeta os resultado que dependem também da complexidade das ações que estão a ser reconhecidas. Conseguímos obter melhores resultados que algumas das abordagens existentes na literatura em reconhecimento de atividade, efetuamos o reconhecimento de forma antecipada e obtivemos resultados encorajadores na previsão de atividades.
Human Activity Recognition is an interdisciplinary research area that has been attracting interest from several research communities specialized in machine learning, computer vision, and medical research. The potential applications range from surveillance systems, human computer interfaces, sports analysis, digital assistants, collaborative robots, health-care and self-driving cars. Capturing human activity presents technical difficulties like occlusion, insufficient lighting, unreliable tracking and ethical concerns. Human motion can be ambiguous and have multiple intents. The complexity of our lives and how we interact with other humans and objects prompt to a nearly infinite combination of variations in how we do things. The focus of this dissertation is to develop a system capable of recognizing and predicting human activity using machine learning techniques to extract meaning from features computed from relevant joints of the human body captured by the skeleton tracker of the Kinect sensor. We propose a modular framework that performs off-line temporal segmentation of sequences of actions, off-line semi unsupervised labeling of sub-activities via clustering techniques, real-time frame by-frame sub-activity recognition using random decision forest binary classifiers right from the very first frames of the action and real-time activity prediction with conditional random fields to model the sequential structure of sequences of actions to reason about future possibilities. We recorded a new dataset containing long sequences of aggressive actions with a total of 72 sequences, 360 samples of 8 distinct actions performed by 12 subjects. We experimented extensively with two different datasets, compared the recognition performance of several supervised classifiers trained with manually labeled data versus semi-unsupervised labeled data. We learned how the quality of the training data affects the results which also depends on the complexity of the actions being recognized. We outperformed state-ofthe-art activity recognition approaches, performed early action recognition and obtained encouraging results in activity prediction.

abstract: The human motion is defined as an amalgamation of several physical traits such as bipedal locomotion, posture and manual dexterity, and mental expectation. In addition to the “positive” body form defined by these traits, casting light on the body produces a “negative” of the body: its shadow. We often interchangeably use with silhouettes in the place of shadow to emphasize indifference to interior features. In a manner of speaking, the shadow is an alter ego that imitates the individual. The principal value of shadow is its non-invasive behaviour of reflecting precisely the actions of the individual it is attached to. Nonetheless we can still think of the body’s shadow not as the body but its alter ego. Based on this premise, my thesis creates an experiential system that extracts the data related to the contour of your human shape and gives it a texture and life of its own, so as to emulate your movements and postures, and to be your extension. In technical terms, my thesis extracts abstraction from a pre-indexed database that could be generated from an offline data set or in real time to complement these actions of a user in front of a low-cost optical motion capture device like the Microsoft Kinect. This notion could be the system’s interpretation of the action which creates modularized art through the abstraction’s ‘similarity’ to the live action. Through my research, I have developed a stable system that tackles various connotations associated with shadows and the need to determine the ideal features that contribute to the relevance of the actions performed. The implication of Factor Oracle [3] pattern interpretation is tested with a feature bin of videos. The system also is flexible towards several methods of Nearest Neighbours searches and a machine learning module to derive the same output. The overall purpose is to establish this in real time and provide a constant feedback to the user. This can be expanded to handle larger dynamic data. In addition to estimating human actions, my thesis best tries to test various Nearest Neighbour search methods in real time depending upon the data stream. This provides a basis to understand varying parameters that complement human activity recognition and feature matching in real time.
Dissertation/Thesis
Masters Thesis Electrical Engineering 2016

Falling in indoor home setting can be dangerous for elderly population (in USA and globally), causing hospitalization, long term reduced mobility, disability or even death. Prevention of fall by monitoring different human activities or identifying the aftermath of fall has greater significance for elderly population. This is possible due to the availability and emergence of miniaturized sensors with advanced electronics and data analytics tools. This thesis aims at developing machine learning models to classify fall activities and non-fall activities. In this thesis, two types of neural networks with different parameters were tested for their capability in dealing with such tasks. A publicly available dataset was used to conduct the experiments. The two types of neural network models, convolution and recurrent neural network, were developed and evaluated. Convolution neural network achieved an accuracy of over 95% for classifying fall and non-fall activities. Recurrent neural network provided an accuracy of over 97% accuracy in predicting fall, non-fall and a third category activity (defined in this study as “pre/postcondition”). Both neural network models show high potential for being used in fall prevention and management activity. Moreover, two theoretical designs of fall detection systems were proposed in this thesis based on the developed convolution and recurrent neural networks.

碩士
國立東華大學
化學系
98
P-glycoprotein (P-gp) is an ATP-dependent membrane transporter that plays a pivotal role in eliminating xenobiotics by active extrusion of xenobiotics from the cell. Multidrug resistance (MDR) is highly associated with the overexpression of P-glycoprotein (P-gp) by cells, resulting in increased efflux of chemotherapeutical agents and reduction of intracellular drug accumulation. It is of clinical importance to develop P-gp inhibitors that can reverse MDR in the process of drug discovery and development. An in silico model was derived to predict the inhibition of P-gp using the newly invented pharmacophore ensemble/support vector machine (PhE/SVM) scheme based on the data compiled from the literature. The predictions by the PhE/SVM model are in good agreement with the experimental observations for those molecules in the training set (n = 31, r2 = 0.89, q2 = 0.86, RMSE=0.40, s = 0.28), the test set (n = 88, r2 = 0.87, RMSE = 0.39, s = 0.25) and the outlier set (n = 10, r2 = 0.96, RMSE = 0.10, s = 0.05). The generated PhE/SVM model also showed high accuracy when subjected to those validation criteria generally adopted to gauge the predictivity of a theoretical model. Thus, it can be asserted that this PhE/SVM model is an accurate, fast and robust model and can be employed to predict P-gp inhibitor binding affinity to facilitate drug discovery and drug development by designing drug candidates with better metabolism profile.

Angiotensin II (AngII), the major effector of the renin angiotensin system, mediates kidney disease progression by signalling through AT-1 receptor (AT-1R), but there are no specific measures of renal AngII activity. Accordingly, we sought to define an AngII-regulated proteome in primary human proximal tubular cells (PTEC) in order to identify potential AngII activity markers in the kidney. We utilized stable isotope labelling with amino acids (SILAC) in PTECs to compare proteomes of AngII-treated and control cells. Of 4618 quantified proteins, 83 were differentially regulated. SILAC ratios for 18 candidates were confirmed by Selected Reaction Monitoring (SRM) assays. Both SILAC and SRM revealed the nuclear factor erythroid 2-related 2 (Nrf2) target protein, heme oxygenase-1 (HO-1) as the most significantly upregulated protein in response to AngII stimulation. AngII-dependent regulation of HO-1 gene and protein was further verified by qRT-PCR and ELISA in PTECs. In order to extend these in vitro observations, we utilized a systems biology approach. We thus overlaid a network of significantly enriched gene ontology (GO) terms from our AngII-regulated proteins with a dataset of differentially expressed kidney genes from AngII-treated wild type mice and AT-1R knock-out mice. Five GO terms were enriched both in vitro and in vivo, and all included HO-1. Furthermore, four additional Nrf2 target proteins were functionally important in vitro and in vivo. We then studied HO-1 kidney expression and urinary excretion in AngII-treated wild type mice and mice with PTEC-specific AT-1R gene deletion. Deletion of the AT-1R gene in PTECs lowered both kidney expression and urine excretion of HO-1, confirming AngII/AT-1R mediated regulation of HO-1. In summary, our in vitro experiments identified novel molecular markers of AngII activity in PTECs and the animal studies demonstrated that these markers also reflect AngII activity in PTECs in vivo. These interesting proteins hold promise as specific markers of renal AngII activity in patients and in experimental models.

abstract: Behavioral health problems such as physical inactivity are among the main causes of mortality around the world. Mobile and wireless health (mHealth) interventions offer the opportunity for applying control engineering concepts in behavioral change settings. Social Cognitive Theory (SCT) is among the most influential theories of health behavior and has been used as the conceptual basis of many behavioral interventions. This dissertation examines adaptive behavioral interventions for physical inactivity problems based on SCT using system identification and control engineering principles. First, a dynamical model of SCT using fluid analogies is developed. The model is used throughout the dissertation to evaluate system identification approaches and to develop control strategies based on Hybrid Model Predictive Control (HMPC). An initial system identification informative experiment is designed to obtain basic insights about the system. Based on the informative experimental results, a second optimized experiment is developed as the solution of a formal constrained optimization problem. The concept of Identification Test Monitoring (ITM) is developed for determining experimental duration and adjustments to the input signals in real time. ITM relies on deterministic signals, such as multisines, and uncertainty regions resulting from frequency domain transfer function estimation that is performed during experimental execution. ITM is motivated by practical considerations in behavioral interventions; however, a generalized approach is presented for broad-based multivariable application settings such as process control. Stopping criteria for the experimental test utilizing ITM are developed using both open-loop and robust control considerations. A closed-loop intensively adaptive intervention for physical activity is proposed relying on a controller formulation based on HMPC. The discrete and logical features of HMPC naturally address the categorical nature of the intervention components that include behavioral goals and reward points. The intervention incorporates online controller reconfiguration to manage the transition between the behavioral initiation and maintenance training stages. Simulation results are presented to illustrate the performance of the system using a model for a hypothetical participant under realistic conditions that include uncertainty. The contributions of this dissertation can ultimately impact novel applications of cyberphysical system in medical applications.
Dissertation/Thesis
Doctoral Dissertation Electrical Engineering 2016

The utilization of in vitro tests with a tiered testing strategy for detection of mild ocular irritants can reduce the use of animals for testing, provide mechanistic data on toxic effects, and reduce the uncertainty associated with dose selection for clinical trials. The first section of this thesis describes how in vitro methods can be used to improve the prediction of the toxicity of chemicals and ophthalmic products. The proper utilization of in vitro methods can accurately predict toxic threshold levels and reduce animal use in product development. Sections two, three and four describe the development of new sensitive in vitro methods for predicting ocular toxicity. Maintaining the barrier function of the cornea is critical for the prevention of the penetration of infections microorganisms and irritating chemicals into the eye. Chapter 2 describes the development of a method for assessing the effects of chemicals on tight junctions using a human corneal epithelial and canine kidney epithelial cell line. In Chapter 3 a method that uses a primary organ culture for assessing single instillation and multiple instillation toxic effects is described. The ScanTox system was shown to be an ideal system to monitor the toxic effects over time as multiple readings can be taken of treated bovine lenses using the nondestructive method of assessing for the lens optical quality. Confirmations of toxic effects were made with the utilization of the viability dye alamarBlue. Chapter 4 describes the development of sensitive in vitro assays for detecting ocular toxicity by measuring the effects of chemicals on the mitochondrial integrity of bovine cornea, bovine lens epithelium and corneal epithelial cells, using fluorescent dyes. The goal of this research was to develop an in vitro test battery that can be used to accurately predict the ocular toxicity of new chemicals and ophthalmic formulations. By comparing the toxicity seen in vivo animals and humans with the toxicity response in these new in vitro methods, it was demonstrated that these in vitro methods can be utilized in a tiered testing strategy in the development of new chemicals and ophthalmic formulations.