ABSTRACT
The recent developments in light generation and detection techniques have opened new possibilities for optical medical imaging, tomography, and diagnosis at tissue penetration depths of ~10 cm. However, because light scattering and diffusion in biological tissue are rather strong, the reconstruction of object images from optical projections needs special attention. We describe a simple reconstruction method for diffuse optical imaging, based on a modified backprojection approach for medical tomography. Specifically, we have modified the standard backprojection method commonly used in x-ray tomographic imaging to include the effects of both the diffusion and the scattering of light and the associated nonlinearities in projection image formation. These modifications are based primarily on the deconvolution of the broadened image by a spatially variant point-spread function that is dependent on the scattering of light in tissue. The spatial dependence of the deconvolution and nonlinearity corrections for the curved propagating ray paths in heterogeneous tissue are handled semiempirically by coordinate transformations. We have applied this method to both theoretical and experimental projections taken by parallel- and fan-beam tomography geometries. The experimental objects were biomedical phantoms with multiple objects, including in vitro animal tissue. The overall results presented demonstrate that image-resolution improvements by nearly an order of magnitude can be obtained. We believe that the tomographic method presented here can provide a basis for rapid, real-time medical monitoring by the use of optical projections. It is expected that such optical tomography techniques can be combined with the optical tissue diagnosis methods based on spectroscopic molecular signatures to result in a versatile optical diagnosis and imaging technology.
ABSTRACT
A new optical neural-network concept using the control of the modes of an injection laser by external feedback is described by a simple laser model. This approach uses the wavelength dispersed longitudinal modes of the laser as neurons and the amount of external feedback as connection weights. The predictions of the simple model are confirmed both with extensive numerical examples using the laser rate equations and also by experiments with GaAlAs injection lasers. The inputs and connection weights to this laser neural network are provided by external masks which control the amount of feedback reaching the laser. Stochastic learning is used to obtain weight masks for a small three-input and four-output neural net for the numerical and experimental examples. Winner-take-all and exclusive-or operations are obtained on the input set with different weight masks. Both of these operations are also obtained in experiments with a three-input/four-output laser neural network operating at an estimated speed greater than 10 GCPS. The eventual speed of this type of neural network hardware is expected to reach well within TCPS range if it is built in an optoelectronic integrated circuit with dimensions in the order of a mm. Different neural-network architectures possible with this approach are discussed.
ABSTRACT
Describes a new approach for obtaining neural network functionality using fully distributed electronic transport rather than lumped electronic circuit elements. For this, vector mapping abilities of a two-dimensional nonlinear inhomogeneous layer are analyzed. This layer is modeled as an inhomogeneous inversion layer in a multiterminal field effect semiconductor device. The author gives computed results as examples of nonlinear vector mapping abilities including nontrivial logic functions with such a layer. These results are achieved by defining relative or differential output signals for the representation of the output information. The type of mapping achieved here is analogous to the one with high-order neural networks. The memory function in the author's structure is imbedded in the distribution of the inhomogeneities.
