The positron emission mammography/tomography breast imaging and biopsy system (PEM/PET): design, construction and phantom-based measurements.

Tomographic breast imaging techniques can potentially improve detection and diagnosis of cancer in women with radiodense and/or fibrocystic breasts. We have developed a high-resolution positron emission mammography/tomography imaging and biopsy device (called PEM/PET) to detect and guide the biopsy of suspicious breast lesions. PET images are acquired to detect suspicious focal uptake of the radiotracer and guide biopsy of the area. Limited-angle PEM images could then be used to verify the biopsy needle position prior to tissue sampling. The PEM/PET scanner consists of two sets of rotating planar detector heads. Each detector consists of a 4 x 3 array of Hamamatsu H8500 flat panel position sensitive photomultipliers (PSPMTs) coupled to a 96 x 72 array of 2 x 2 x 15 mm(3) LYSO detector elements (pitch = 2.1 mm). Image reconstruction is performed with a three-dimensional, ordered set expectation maximization (OSEM) algorithm parallelized to run on a multi-processor computer system. The reconstructed field of view (FOV) is 15 x 15 x 15 cm(3). Initial phantom-based testing of the device is focusing upon its PET imaging capabilities. Specifically, spatial resolution and detection sensitivity were assessed. The results from these measurements yielded a spatial resolution at the center of the FOV of 2.01 +/- 0.09 mm (radial), 2.04 +/- 0.08 mm (tangential) and 1.84 +/- 0.07 mm (axial). At a radius of 7 cm from the center of the scanner, the results were 2.11 +/- 0.08 mm (radial), 2.16 +/- 0.07 mm (tangential) and 1.87 +/- 0.08 mm (axial). Maximum system detection sensitivity of the scanner is 488.9 kcps microCi(-1) ml(-1) (6.88%). These promising findings indicate that PEM/PET may be an effective system for the detection and diagnosis of breast cancer.

Parameter incremental learning algorithm for neural networks.

In this paper, a novel stochastic (or online) training algorithm for neural networks, named parameter incremental learning (PIL) algorithm, is proposed and developed. The main idea of the PIL strategy is that the learning algorithm should not only adapt to the newly presented input-output training pattern by adjusting parameters, but also preserve the prior results. A general PIL algorithm for feedforward neural networks is accordingly presented as the first-order approximate solution to an optimization problem, where the performance index is the combination of proper measures of preservation and adaptation. The PIL algorithms for the multilayer perceptron (MLP) are subsequently derived. Numerical studies show that for all the three benchmark problems used in this paper the PIL algorithm for MLP is measurably superior to the standard online backpropagation (BP) algorithm and the stochastic diagonal Levenberg-Marquardt (SDLM) algorithm in terms of the convergence speed and accuracy. Other appealing features of the PIL algorithm are that it is computationally as simple as the BP algorithm, and as easy to use as the BP algorithm. It, therefore, can be applied, with better performance, to any situations where the standard online BP algorithm is applicable.