SEARCH · PubMed Health

Results for “multimodal artificial intelligence”

Explore indexed PubMed citations for clinical trials, systematic reviews and public health research. Read source abstracts and follow each citation to its original PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

79 records · Page 5Linked to original sources

PubMed2004

On a multimode test sequencing problem.

Test sequencing is a binary identification problem wherein one needs to develop a minimal expected cost test procedure to determine which one of a finite number of possible failure states, if any, is present. In this paper, we consider a multimode test sequencing (MMTS) problem, in which tests are distributed among multiple modes and additional transition costs will be incurred if a test sequence involves mode changes. The multimode test sequencing problem can be solved optimally via dynamic programming or AND/OR graph search methods. However, for large systems, the associated computation with dynamic programming or AND/OR graph search methods is substantial due to the rapidly increasing number of OR nodes (denoting ambiguity states and current modes) and AND nodes (denoting next modes and tests) in the search graph. In order to overcome the computational explosion, we propose to apply three heuristic algorithms based on information gain: information gain heuristic (IG), mode capability evaluation (MC), and mode capability evaluation with limited exploration of depth and degree of mode Isolation (MCLEI). We also propose to apply rollout strategies, which are guaranteed to improve the performance of heuristics, as long as the heuristics are sequentially improving. We show computational results, which suggest that the information-heuristic based rollout policies are significantly better than traditional information gain heuristic. We also show that among the three information heuristics proposed, MCLEI achieves the best tradeoff between optimality and computational complexity.

Algorithms↗

PubMedDate not supplied

Toward the discrimination of early melanoma from common and dysplastic nevus using fiber optic diffuse reflectance spectroscopy.

We describe a study of the discrimination of early melanoma from common and dysplastic nevus using fiber optic diffuse reflectance spectroscopy. Diffuse reflectance spectra in the wavelength range 550 to 1000 nm are obtained using 400-microm core multimode fibers arranged in a six-illumination-around-one-collection geometry with a single fiber-fiber spacing of 470 microm. Spectra are collected at specific locations on 120 pigmented lesions selected by clinicians as possible melanoma, including 64 histopathologically diagnosed as melanoma. These locations are carried through to the histopathological diagnosis, permitting a spatially localized comparison with the corresponding spectrum. The variations in spectra between groups of lesions with different diagnoses are examined and reduced to features suitable for discriminant analysis. A classifier distinguishing between benign and malignant lesions performs with sensitivity/specificity of between 6469% and 7278%. Classifiers between pairs of the group common nevus, dysplastic nevus, in situ melanoma, and invasive melanoma show better or similar performance than the benign/malignant classifier, and analysis provides evidence that different spectral features are needed for each pair of groups. This indicates that multiple discriminant systems are likely to be required to distinguish between melanoma and similar lesions.

Algorithms↗

PubMed1997

Simulation of soft-tissue tumor excisions: a multimodal interactive approach.

Total 3-D reconstruction of the tumor size, shape, and relations with surrounding structures using CT, MRI, sonography, and angiography images can make simulated radical resection of soft-tissue sarcomas possible, thus sparing normal tissues. With our approach, starting from three MR images for a given patient, a new single image representation of all three parameters is generated by using two different techniques on a workstation in a standard UNIX and X-11 environment. The first one is a transformation linking together the MR parameters and the RGB (red, green, blue) color components. The second one is an unsupervised segmentation method based on a number of neural and fuzzy models. We can dinamically render and update a stereo display using field sequential presentation of left and right eye views on the monitor, with Cristal Eyes LCD shutter eyewear (StereoGraphics Inc., San Rafael, CA) to view it. As 3D locating tool, a 3D locating control system based on low-frequency magnetic fields (Polhemus Fastrak) has been chosen. Simulations of soft-tissues excisions may be performed in this interactive environment with augmented-reality modalities. All this, in our experience, has greatly facilitated the simulation of soft-tissue sarcoma excisions.

Artificial Intelligence↗

PubMed2004

Combination of hyperbolic functions for multimodal biometrics data fusion.

In this paper, we treat the problem of combining fingerprint and speech biometric decisions as a classifier fusion problem. By exploiting the specialist capabilities of each classifier, a combined classifier may yield results which would not be possible in a single classifier. The Feedforward Neural Network provides a natural choice for such data fusion as it has been shown to be a universal approximator. However, the training process remains much to be a trial-and-error effort since no learning algorithm can guarantee convergence to optimal solution within finite iterations. In this work, we propose a network model to generate different combinations of the hyperbolic functions to achieve some approximation and classification properties. This is to circumvent the iterative training problem as seen in neural networks learning. In many decision data fusion applications, since individual classifiers or estimators to be combined would have attained a certain level of classification or approximation accuracy, this hyperbolic functions network can be used to combine these classifiers taking their decision outputs as the inputs to the network. The proposed hyperbolic functions network model is first applied to a function approximation problem to illustrate its approximation capability. This is followed by some case studies on pattern classification problems. The model is finally applied to combine the fingerprint and speaker verification decisions which show either better or comparable results with respect to several commonly used methods.

Algorithms↗

PubMed1997

Brainvox: an interactive, multimodal visualization and analysis system for neuroanatomical imaging.

A study of cognition emerging from a neurobiological perspective, as opposed to one emerging from a purely computational or psychological perspective, begins with observations of the human brain in normal and pathological states and is furthered by the investigation of hypotheses which are articulated using neuroanatomical nomenclature. Brainvox is an interactive three-dimensional brain imaging software package designed to permit such research through the support of the description and quantification of brain pathology in magnetic resonance images and of the experimental investigation of human cognition in lesion and functional imaging studies. Important general features of Brainvox, for these purposes, are: (1) adaptation of volume rendering for brain lesions and for corendered datasets; (2) shared memory architecture, which enables the user to identify and label anatomical structures, while inspecting the brain in multiple views simultaneously; (3) modular program design, including interlocking command-line utilities, which make Brainvox extensible and empower users without programming expertise to implement new analysis techniques through Unix shell scripting; and (4) full integration of three-dimensional tools for visualization with tools for analysis. Specific features include a new object templating technique (MAP-3) for studies of groups of brain-lesioned subjects, a complete and extensible suite of command-line processing utilities, a three-dimensional optimal graph-searching tool, and a method for planning PET slices and matching MR and PET slices (MP_FIT).

Artificial Intelligence↗

PubMed2005

Locally linear discriminant analysis for multimodally distributed classes for face recognition with a single model image.

We present a novel method of nonlinear discriminant analysis involving a set of locally linear transformations called "Locally Linear Discriminant Analysis (LLDA)." The underlying idea is that global nonlinear data structures are locally linear and local structures can be linearly aligned. Input vectors are projected into each local feature space by linear transformations found to yield locally linearly transformed classes that maximize the between-class covariance while minimizing the within-class covariance. In face recognition, linear discriminant analysis (LDA) has been widely adopted owing to its efficiency, but it does not capture nonlinear manifolds of faces which exhibit pose variations. Conventional nonlinear classification methods based on kernels such as generalized discriminant analysis (GDA) and support vector machine (SVM) have been developed to overcome the shortcomings of the linear method, but they have the drawback of high computational cost of classification and overfitting. Our method is for multiclass nonlinear discrimination and it is computationally highly efficient as compared to GDA. The method does not suffer from overfitting by virtue of the linear base structure of the solution. A novel gradient-based learning algorithm is proposed for finding the optimal set of local linear bases. The optimization does not exhibit a local-maxima problem. The transformation functions facilitate robust face recognition in a low-dimensional subspace, under pose variations, using a single model image. The classification results are given for both synthetic and real face data.

Algorithms↗

PubMed2003

[Prediction of postoperative nausea and vomiting using an artificial neural network].

OBJECTIVE: Postoperative nausea and vomiting (PONV) are still frequent side-effects after general anaesthesia. These unpleasant symptoms for the patients can be sufficiently reduced using a multimodal antiemetic approach. However, these efforts should be restricted to risk patients for PONV. Thus, predictive models are required to identify these patients before surgery. So far all risk scores to predict PONV are based on results of logistic regression analysis. Artificial neural networks (ANN) can also be used for prediction since they can take into account complex and non-linear relationships between predictive variables and the dependent item. This study presents the development of an ANN to predict PONV and compares its performance with two established simplified risk scores (Apfel's and Koivuranta's scores). METHODS: The development of the ANN was based on data from 1,764 patients undergoing elective surgical procedures under balanced anaesthesia. The ANN was trained with 1,364 datasets and a further 400 were used for supervising the learning process. One of the 49 ANNs showing the best predictive performance was compared with the established risk scores with respect to practicability, discrimination (by means of the area under a receiver operating characteristics curve) and calibration properties (by means of a weighted linear regression between the predicted and the actual incidences of PONV). RESULTS: The ANN tested showed a statistically significant ( p<0.0001) and clinically relevant higher discriminating power (0.74; 95% confidence interval: 0.70-0.78) than the Apfel score (0.66; 95% CI: 0.61-0.71) or Koivuranta's score (0.69; 95% CI: 0.65-0.74). Furthermore, the agreement between the actual incidences of PONV and those predicted by the ANN was also better and near to an ideal fit, represented by the equation y=1.0x+0. The equations for the calibration curves were: KNN y=1.11x+0, Apfel y=0.71x+1, Koivuranta 0.86x-5. CONCLUSION: The improved predictive accuracy achieved by the ANN is clinically relevant. However, the disadvantages of this system prevail because a computer is required for risk calculation. Thus, we still recommend the use of one of the simplified risk scores for clinical practice.

Artificial Intelligence↗