Pub Date : 1900-01-01DOI: 10.1109/NNSP.2003.1318052
P. Sanguansat, P. Yanwit, P. Tangwiwatwong, W. Asdornwised, S. Jitapunkul
We propose a new method for recognition - the domain-dependent bilingual hand-printed character recognition. We implemented two-stage recognition systems based on two important character properties, defined as spatial and temporal informative features. The proposed spatial informative features (SIF) are off-line characters' structures that are exploited in order to differentiate Thai from English characters. These features can also be called distinctive features (DF). In contrast, temporal informative features (TIF) are segments of characters extracted using our proposed features, called start-to-end point distance feature, and other standard on-line features. Our proposed TIF features help us to solve ambiguity occurred in several Thai and English character, which conventional features cannot resolve. In the recognition system, the first stage is performed the language classification task using distinctive features, while the second stage is using hidden Markov model (HMM) as the final classifier. The advantages of using language classification at the first recognition stage are two folds. First, the decision complexity at the final recognition stage can be reduced. Second, the observation stages of two independent language HMMs can be better optimized than one bilingual HMM. From the experimental results, language classification recognition accuracy is 99.31%, while the recognition accuracy of Thai and English characters are 91.67% and 90.23%, respectively. Hence, the overall recognition accuracy is 91.05%.
{"title":"Language-based hand-printed character recognition: a novel method using spatial and temporal informative features","authors":"P. Sanguansat, P. Yanwit, P. Tangwiwatwong, W. Asdornwised, S. Jitapunkul","doi":"10.1109/NNSP.2003.1318052","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318052","url":null,"abstract":"We propose a new method for recognition - the domain-dependent bilingual hand-printed character recognition. We implemented two-stage recognition systems based on two important character properties, defined as spatial and temporal informative features. The proposed spatial informative features (SIF) are off-line characters' structures that are exploited in order to differentiate Thai from English characters. These features can also be called distinctive features (DF). In contrast, temporal informative features (TIF) are segments of characters extracted using our proposed features, called start-to-end point distance feature, and other standard on-line features. Our proposed TIF features help us to solve ambiguity occurred in several Thai and English character, which conventional features cannot resolve. In the recognition system, the first stage is performed the language classification task using distinctive features, while the second stage is using hidden Markov model (HMM) as the final classifier. The advantages of using language classification at the first recognition stage are two folds. First, the decision complexity at the final recognition stage can be reduced. Second, the observation stages of two independent language HMMs can be better optimized than one bilingual HMM. From the experimental results, language classification recognition accuracy is 99.31%, while the recognition accuracy of Thai and English characters are 91.67% and 90.23%, respectively. Hence, the overall recognition accuracy is 91.05%.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130327822","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.1109/NNSP.2003.1318008
D. Muramatsu, S. Hashimoto, T. Tsunashima, T. Kaburagi, M. Sasaki, Takashi Matsumoto
A new algorithm is proposed for inferring the number of transmembrane regions of transmembrane proteins from two dimensional vector trajectories consisting of hydropathy index and charge of amino acids by stochastic dynamical system models. The prediction accuracy of a preliminary experiment is 94%. Since no fine-tuning is done, this appears encouraging.
{"title":"Inferring transmembrane region counts with hydropathy index/charge two dimensional trajectories of stochastic dynamical systems","authors":"D. Muramatsu, S. Hashimoto, T. Tsunashima, T. Kaburagi, M. Sasaki, Takashi Matsumoto","doi":"10.1109/NNSP.2003.1318008","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318008","url":null,"abstract":"A new algorithm is proposed for inferring the number of transmembrane regions of transmembrane proteins from two dimensional vector trajectories consisting of hydropathy index and charge of amino acids by stochastic dynamical system models. The prediction accuracy of a preliminary experiment is 94%. Since no fine-tuning is done, this appears encouraging.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129095413","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.1109/NNSP.2003.1318085
F. Chereau, I. Defée
This work investigates signal processing networks in which randomness is an inherent feature like in biological neuronal networks. Signal processing operations are usually performed with algorithms requiring high-precision and order. It is thus interesting to investigate how signal processing operations could be realized in systems with inherent randomness which is apparent in neuronal networks. We are studying possible implementation of convolution and correlation operations based on generalized transform approach with rectangular matrices generated by random sequences. Conditions are formulated and illustrated how correlation and convolution operators can be computed with such matrices. We show next that increasing the size of matrices allows to decrease the precision of operations and to introduce substantial quantization and thresholding. The use of random matrices provides also for strong robustness to noise resulting from unreliable operation. We show also that the nonlinearity due to the quantization and thresholding leads naturally to the decorrelation of transformation vectors which might be useful for associative storage.
{"title":"Implementation of signal processing operations by transforms with random coefficients for neuronal systems modelling","authors":"F. Chereau, I. Defée","doi":"10.1109/NNSP.2003.1318085","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318085","url":null,"abstract":"This work investigates signal processing networks in which randomness is an inherent feature like in biological neuronal networks. Signal processing operations are usually performed with algorithms requiring high-precision and order. It is thus interesting to investigate how signal processing operations could be realized in systems with inherent randomness which is apparent in neuronal networks. We are studying possible implementation of convolution and correlation operations based on generalized transform approach with rectangular matrices generated by random sequences. Conditions are formulated and illustrated how correlation and convolution operators can be computed with such matrices. We show next that increasing the size of matrices allows to decrease the precision of operations and to introduce substantial quantization and thresholding. The use of random matrices provides also for strong robustness to noise resulting from unreliable operation. We show also that the nonlinearity due to the quantization and thresholding leads naturally to the decorrelation of transformation vectors which might be useful for associative storage.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"97 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127095936","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.1109/NNSP.2003.1318078
E. Teufel, M. Kletting, W. Teich, H. Pfleiderer, C. Tarin-Sauer
Type-I diabetes mellitus patients can not produce the hormone insulin endogenously. As this hormone is necessary to control the blood sugar level, which is raised by eating, insulin must be delivered exogeneously. Delivering insulin exogeneously demands correct dosage to avoid an extremely high or low blood glucose level. Most patients are not able to administer the adequate insulin dose because they are not able to predict the evolution of their own glucose level after a meal. Therefore, a model of the glucose metabolism is of crucial interest to help patients to determine correct insulin doses. These models shall be capable of predicting the course of the blood glucose level for a couple of hours with reasonable precision. In this paper a computer aided assistance system for diabetes patients running on a mobile handheld device is presented. This assistance system mainly consists of a model of the glucose metabolism, implemented by a modified Elman net. The training is performed through the BPTT algorithm where the training data were generated with an analytical non-linear glucose metabolism model that is quite realistic but cannot be adapted to every single patient. The glucose metabolism process is defined by two inputs, injected insulin and ingested glucose, and one output, namely the blood glucose. Due to the fact that metabolic processes in general have large time constants this process is characterized by the fact that the current net output, that is the blood glucose level, heavily depends on data that are not present in the current input layer any more. The Elman net's context-layer is capable of storing this information. Simulation results demonstrate that the output of this type of neural network closely follows the reference.
{"title":"Modelling the glucose metabolism with backpropagation through time trained Elman nets","authors":"E. Teufel, M. Kletting, W. Teich, H. Pfleiderer, C. Tarin-Sauer","doi":"10.1109/NNSP.2003.1318078","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318078","url":null,"abstract":"Type-I diabetes mellitus patients can not produce the hormone insulin endogenously. As this hormone is necessary to control the blood sugar level, which is raised by eating, insulin must be delivered exogeneously. Delivering insulin exogeneously demands correct dosage to avoid an extremely high or low blood glucose level. Most patients are not able to administer the adequate insulin dose because they are not able to predict the evolution of their own glucose level after a meal. Therefore, a model of the glucose metabolism is of crucial interest to help patients to determine correct insulin doses. These models shall be capable of predicting the course of the blood glucose level for a couple of hours with reasonable precision. In this paper a computer aided assistance system for diabetes patients running on a mobile handheld device is presented. This assistance system mainly consists of a model of the glucose metabolism, implemented by a modified Elman net. The training is performed through the BPTT algorithm where the training data were generated with an analytical non-linear glucose metabolism model that is quite realistic but cannot be adapted to every single patient. The glucose metabolism process is defined by two inputs, injected insulin and ingested glucose, and one output, namely the blood glucose. Due to the fact that metabolic processes in general have large time constants this process is characterized by the fact that the current net output, that is the blood glucose level, heavily depends on data that are not present in the current input layer any more. The Elman net's context-layer is capable of storing this information. Simulation results demonstrate that the output of this type of neural network closely follows the reference.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121788236","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.1109/NNSP.2003.1318048
David J. Miller, J. Browning
Several authors have addressed learning as a classifier given by a mixed labeled/unlabeled training set. These works assumes the unlabeled sample originates from one of the (known) classes. This work considers the scenario in which unlabeled points may belong either to known/predefined or to here-to-fore undiscovered classes. There are several practical situations where such data may arise. We earlier proposed a novel statistical mixture model to fit in this mixed data. In this paper we review the method and introduce an alternative model. Our fundamental strategy is to view as observed the data not only the feature vector and the class label, but also the fact of label presence/absence for each point. Two types of mixture components are used to explain label presence/absence. "Predefined" components generate both labeled and unlabeled points and assume the labels that are missing at random. These components represent the known classes. "Non-predefined" components only generate unlabeled points. In localized regions, the data subsets are captured exclusively unlabeled. Such subsets may represent an outlier distribution, or new classes. The components' predefined/non-predefined natures are data-driven, learned with the other parameters via an algorithm based on expectation-maximization (EM). There are three natural applications presented: 1) robust classifier design, given by a mixed training set with outliers; 2) classification with rejections; and 3) identification of the unlabeled points (and their representative components) originated from unknown classes, i.e. new class discovery. The effectiveness of our models in discovering purely unlabeled data components (potential new classes) is evaluated both by synthetic and real data sets. Although each of our models has its own advantages, the original model is found is achieved by the best class discovery results.
{"title":"A mixture model framework for class discovery and outlier detection in mixed labeled/unlabeled data sets","authors":"David J. Miller, J. Browning","doi":"10.1109/NNSP.2003.1318048","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318048","url":null,"abstract":"Several authors have addressed learning as a classifier given by a mixed labeled/unlabeled training set. These works assumes the unlabeled sample originates from one of the (known) classes. This work considers the scenario in which unlabeled points may belong either to known/predefined or to here-to-fore undiscovered classes. There are several practical situations where such data may arise. We earlier proposed a novel statistical mixture model to fit in this mixed data. In this paper we review the method and introduce an alternative model. Our fundamental strategy is to view as observed the data not only the feature vector and the class label, but also the fact of label presence/absence for each point. Two types of mixture components are used to explain label presence/absence. \"Predefined\" components generate both labeled and unlabeled points and assume the labels that are missing at random. These components represent the known classes. \"Non-predefined\" components only generate unlabeled points. In localized regions, the data subsets are captured exclusively unlabeled. Such subsets may represent an outlier distribution, or new classes. The components' predefined/non-predefined natures are data-driven, learned with the other parameters via an algorithm based on expectation-maximization (EM). There are three natural applications presented: 1) robust classifier design, given by a mixed training set with outliers; 2) classification with rejections; and 3) identification of the unlabeled points (and their representative components) originated from unknown classes, i.e. new class discovery. The effectiveness of our models in discovering purely unlabeled data components (potential new classes) is evaluated both by synthetic and real data sets. Although each of our models has its own advantages, the original model is found is achieved by the best class discovery results.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133466925","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.1109/NNSP.2003.1318081
A. Hegde, Deniz Erdoğmuş, Y. Rao, J. Príncipe, Jianbo Gao
This work addresses the issue of quantifying asymmetric functional relationships between signals. We specifically consider a previously proposed similarity index that is conceptually powerful, yet computationally very expensive. The complexity increases with the square of the number of samples in the signals. In order to counter this difficulty, a self-organizing map that is trained to model the statistical distribution of the signals of interest is introduced in the similarity index evaluation procedure. The SOM based technique is equally accurate, but computationally less expensive compared to the conventional measure. These results are demonstrated by comparing the original and SOM-based similarity index approaches on synthetic chaotic signal and real EEG signal mixtures.
{"title":"SOM-based similarity index measure: quantifying interactions between multivariate structures","authors":"A. Hegde, Deniz Erdoğmuş, Y. Rao, J. Príncipe, Jianbo Gao","doi":"10.1109/NNSP.2003.1318081","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318081","url":null,"abstract":"This work addresses the issue of quantifying asymmetric functional relationships between signals. We specifically consider a previously proposed similarity index that is conceptually powerful, yet computationally very expensive. The complexity increases with the square of the number of samples in the signals. In order to counter this difficulty, a self-organizing map that is trained to model the statistical distribution of the signals of interest is introduced in the similarity index evaluation procedure. The SOM based technique is equally accurate, but computationally less expensive compared to the conventional measure. These results are demonstrated by comparing the original and SOM-based similarity index approaches on synthetic chaotic signal and real EEG signal mixtures.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"64 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114816994","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.1109/NNSP.2003.1318068
Artur J. Ferreira, Mário A. T. Figueiredo
In this paper we address the orthogonalization of independent component analysis (ICA) to obtain transform-based image coders. We consider several classes of training images, from which we extract the independent components, followed by orthogonalization, obtaining bases for image coding. Experimental tests show the generalization ability of ICA of natural images, and the adaptation ability to specific classes. The proposed fixed size block coders have lower transform complexity than JPEG. They outperform JPEG, on several classes of images, for a given range of compression ratios, according to both standard (SNR) and perceptual (picture quality scale - PQS) measures. For some image classes, the visual quality of the images obtained with our coders is similar to that obtained by JPEG2000, which is currently the state of the art still image coder. On fingerprint images, our fixed and variable size block coders perform competitively with the special-purpose wavelet-based coder developed by the FBI.
{"title":"Image compression using orthogonalized independent components bases","authors":"Artur J. Ferreira, Mário A. T. Figueiredo","doi":"10.1109/NNSP.2003.1318068","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318068","url":null,"abstract":"In this paper we address the orthogonalization of independent component analysis (ICA) to obtain transform-based image coders. We consider several classes of training images, from which we extract the independent components, followed by orthogonalization, obtaining bases for image coding. Experimental tests show the generalization ability of ICA of natural images, and the adaptation ability to specific classes. The proposed fixed size block coders have lower transform complexity than JPEG. They outperform JPEG, on several classes of images, for a given range of compression ratios, according to both standard (SNR) and perceptual (picture quality scale - PQS) measures. For some image classes, the visual quality of the images obtained with our coders is similar to that obtained by JPEG2000, which is currently the state of the art still image coder. On fingerprint images, our fixed and variable size block coders perform competitively with the special-purpose wavelet-based coder developed by the FBI.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"127 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133130648","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.1109/NNSP.2003.1318072
T. Jan, M. Piccardi, T. Hintz
In automated visual surveillance applications, detection of suspicious human behaviors is of great practical importance. However due to random nature of human movements, reliable classification of suspicious human movements can be very difficult. Artificial neural network (ANN) classifiers can perform well however their computational requirements can be very large for real time implementation. In this paper, a data-based modeling neural network such as modified probabilistic neural network (MPNN) is introduced which partitions the decision space nonlinearly in order to achieve reliable classification, however still with acceptable computations. The experiment shows that the compact MPNN attains good classification performance compared to that of other larger conventional neural network based classifiers such as multilayer perceptron (MLP) and self organising map (SOM).
{"title":"Neural network classifiers for automated video surveillance","authors":"T. Jan, M. Piccardi, T. Hintz","doi":"10.1109/NNSP.2003.1318072","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318072","url":null,"abstract":"In automated visual surveillance applications, detection of suspicious human behaviors is of great practical importance. However due to random nature of human movements, reliable classification of suspicious human movements can be very difficult. Artificial neural network (ANN) classifiers can perform well however their computational requirements can be very large for real time implementation. In this paper, a data-based modeling neural network such as modified probabilistic neural network (MPNN) is introduced which partitions the decision space nonlinearly in order to achieve reliable classification, however still with acceptable computations. The experiment shows that the compact MPNN attains good classification performance compared to that of other larger conventional neural network based classifiers such as multilayer perceptron (MLP) and self organising map (SOM).","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"69 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127400870","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.1109/NNSP.2003.1318076
P. Gallant, G. Aitken
A genetic algorithm that designs artificial neural networks for time-series prediction encodes the structure and the weight magnitudes in a novel genome representation. This allows the genetic algorithm to perform training and complexity control simultaneously, thus directly addressing the problems of generalization and overfitting of data in the evolution of the network. Modified genetic crossover and modified mutation operations are introduced to increase population diversity and improve speed of convergence. Well performing neural networks were evolved automatically for time-series prediction of atmospherically-perturbed light waves in adaptive optics and the time series used in the 1998 Leuven predictor competition.
{"title":"Genetic algorithm design of complexity-controlled time-series predictors","authors":"P. Gallant, G. Aitken","doi":"10.1109/NNSP.2003.1318076","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318076","url":null,"abstract":"A genetic algorithm that designs artificial neural networks for time-series prediction encodes the structure and the weight magnitudes in a novel genome representation. This allows the genetic algorithm to perform training and complexity control simultaneously, thus directly addressing the problems of generalization and overfitting of data in the evolution of the network. Modified genetic crossover and modified mutation operations are introduced to increase population diversity and improve speed of convergence. Well performing neural networks were evolved automatically for time-series prediction of atmospherically-perturbed light waves in adaptive optics and the time series used in the 1998 Leuven predictor competition.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129100858","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.1109/NNSP.2003.1318070
A. Hadjiprocopis, P. Tofts
Multiple sclerosis (MS) is a chronic disease of the central nervous system which attacks the insulating myelin coating of nerve fibers in the brain and spinal cord, leaving scar tissue which can be seen on magnetic resonance imaging (MRI) scans. There is a well recognised need for a robust, objective, accurate and reproducible automatic method for identifying multiple sclerosis lesions on proton density (PD) and T/sub 2/-weighted MRI. Feed-forward neural networks (FFNN) are computational techniques inspired by the physiology of the brain and used in the approximation of general mappings from one finite dimensional space to another. They present a practical application of the theoretical resolution of Hilbert's 13th problem by Kolmogorov and Lorenz, and have been used with success in a variety of applications. We present a method for automatic MS lesion segmentation for fast spin echo (FSE) images (PD-weighted & T/sub 2/-weighted) based on an ensemble of feed-forward neural networks. The FFNN of the input layer of the ensemble are trained with different portions of example lesion and non-lesion data which have previously been hand-segmented by a clinician. The final output of the ensemble is determined by a gate FFNN which is trained to weigh the response of the input layer to unseen training data. The ensemble was trained with data from 14 MS patients and evaluated with data from another 6. The results are presented.
{"title":"An automatic lesion segmentation method for fast spin echo magnetic resonance images using an ensemble of neural networks","authors":"A. Hadjiprocopis, P. Tofts","doi":"10.1109/NNSP.2003.1318070","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318070","url":null,"abstract":"Multiple sclerosis (MS) is a chronic disease of the central nervous system which attacks the insulating myelin coating of nerve fibers in the brain and spinal cord, leaving scar tissue which can be seen on magnetic resonance imaging (MRI) scans. There is a well recognised need for a robust, objective, accurate and reproducible automatic method for identifying multiple sclerosis lesions on proton density (PD) and T/sub 2/-weighted MRI. Feed-forward neural networks (FFNN) are computational techniques inspired by the physiology of the brain and used in the approximation of general mappings from one finite dimensional space to another. They present a practical application of the theoretical resolution of Hilbert's 13th problem by Kolmogorov and Lorenz, and have been used with success in a variety of applications. We present a method for automatic MS lesion segmentation for fast spin echo (FSE) images (PD-weighted & T/sub 2/-weighted) based on an ensemble of feed-forward neural networks. The FFNN of the input layer of the ensemble are trained with different portions of example lesion and non-lesion data which have previously been hand-segmented by a clinician. The final output of the ensemble is determined by a gate FFNN which is trained to weigh the response of the input layer to unseen training data. The ensemble was trained with data from 14 MS patients and evaluated with data from another 6. The results are presented.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"74 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127645667","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：