Isaac N. Bankman

Optimal detection, classification, and superposition resolution in neural waveform recordings

The effects of noise autocorrelation on neural waveform recognition (detection, classification, and superposition resolution) are investigated, using microelectrode recordings from the cortex of a monkey. Optimal waveform recognition is accomplished by passing the data through a whitening filter before matched filtering for detection or template matching for classification and superposition resolution. Template matching without whitening requires about 40% higher signal-to-noise ratio (SNR) than template matching with whitening for comparable classification and superposition resolution. The comparable difference for detection is 15%.<<ETX>>

Feature-based detection of the K-complex wave in the human electroencephalogram using neural networks

The main difficulties in reliable automated detection of the K-complex wave in EEG are its close similarity to other waves and the lack of specific characterization criteria. The authors present a feature-based detection approach using neural networks that provides good agreement with visual K-complex recognition: a sensitivity of 90% is obtained with about 8% false positives. The respective contribution of the features and that of the neural network is demonstrated by comparing the results to those obtained with (i) raw EEG data presented to neural networks, and (ii) features presented to Fishers linear discriminant.<<ETX>>

Underwater optical communications systems. Part 2: basic design considerations

Acoustic systems may provide suitable underwater communications because sound propagates well in water. However, the maximum data transmission rates of these systems in shallow littoral waters are ~10 kilobits per second (kbps) which may be achieved only at ranges of less than 100 m. Although underwater (u/w) wireless optical communications systems can have even shorter ranges due to greater attenuation of light propagating through water, they may provide higher bandwidth (up to several hundred kbps) communications as well as covertness. To exploit these potential advantages, we consider the basic design issues for u/w optical communications systems in this paper. In addition to the basic physics of u/w optical communications with environmental noise, we consider system performance with some state-of-the-art commercial off-the-shelf (COTS) components, which have promise for placing u/w optical communications systems in a small package with low power consumption and weight. We discuss light sources which show promise for u/w optical transmitters such as laser diodes (LDs) and light emitting diodes (LEDs). Laser diodes with their output frequency shifted into the 500- to 650-nm range can emit more energy per pulse than LEDs but are more expensive. Currently, LEDs emit substantial amounts of light and are typically very inexpensive. Also, COTS photodiodes can be used as detectors which can respond to pulses several nanoseconds wide. Transmitter broadcast angles and detector fields of view (FOVs) with pointing considerations are discussed. If the transmitter broadcast angle and the detector FOV are both narrow, the signal-to-noise ratio (SNR) of the received pulse is higher but the pointing accuracy of transmitter and receiver is critical. If, however, the transmitter broadcast angle and/or the detector FOV is wide, pointing is less critical but SNR is lower and some covertness may be lost. The propagation of the transmitted light in various clear oceanic and turbid coastal water types is considered with range estimates for some COTS light sources and detectors. We also consider the effects of environmental noise such as background solar radiation, which typically limits performance of these systems

Segmentation algorithms for detecting microcalcifications in mammograms

The presence of microcalcification clusters in mammograms contributes evidence for the diagnosis of early stages of breast cancer. In many cases, microcalcifications are subtle and their detection can benefit from an automated system serving as a diagnostic aid. The potential contribution of such a system may become more significant as the number of mammograms screened increases to levels that challenge the capacity of radiology clinics. Many techniques for detecting microcalcifications start with a segmentation algorithm that indicates all candidate structures for the subsequent phases. Most algorithms used to segment microcalcifications have aspects that might raise operational difficulties, such as thresholds or windows that must be selected, or parametric models of the data. We present a new segmentation algorithm and compare it to two other algorithms: the multi-tolerance region-growing algorithm, which operates without the aspects mentioned above, and the active contour model, which has not been applied previously to segment microcalcifications. The new algorithm operates without threshold or window selection or parametric data models, and it is more than an order of magnitude faster than the other two.

Identification of dynamic mechanical parameters of the human chest during manual cadiopulmonary resuscitation

Timely cardiopulmonary resuscitation (CPR) is often unsuccessful. The outcome can be improved by a better understanding of the relationship between the force applied to the sternum and the resulting hemodynamic effects. The first step in this complex chain of interactions is the mechanical response of the chest wall to cyclical compressions. A dynamic mechanical model of the chest response was formulated, and a method of identification of the model parameters based on force, displacement, and acceleration data acquired during cyclical compressions was developed. The elasticity, damping, and equivalent mass of the human chest were estimated with a constrained nonlinear least-mean-square identification technique. The method was validated on data acquired from a test apparatus built for this purpose. The model fit was measured with the normalized chi-square statistic on residuals obtained between recorded force and force predicted by the model. In the analysis of one human chest, the elasticity was found to be nonlinear and statistically different during compression and release.<<ETX>>

Automated sizing of DNA fragments in atomic force microscope images.

Current techniques used to measure lengths of DNA fragments in atomic force microscope (AFM) images require a user to operate interactive software and execute tedious error-prone cursor selections. An algorithm is proposed which provides an automated method for determining DNA fragment lengths from AFM images without interaction from the computer operator (e.g. cursor selections or mouse clicks). The approach utilises image processing techniques tailored to characteristics of AFM images of DNA fragments. The automated measurements have a mean absolute deviation of less than 1 pixel when compared to manual image-based measurements. The DNA length determined from the histogram of calculated lengths is accurate to within 3% of the actual DNA length in solution. For fragments that are 250 base-pairs long, the precision is estimated to be within 17 nm, which is about 20% of the total length. This precision was confirmed when the algorithm easily resolved fragments in one image that differed by only 17 nm. Fragment sizes up to 2000 base-pairs have been tested and successfully sized. This algorithm is being developed as part of a new solid-state DNA sizing technique for applications such as genotyping and construction of physical genome maps.

Automated recognition of microcalcification clusters in mammograms

The widespread and increasing use of mammographic screening for early breast cancer detection is placing a significant strain on clinical radiologists. Large numbers of radiographic films have to be visually interpreted in fine detail to determine the subtle hallmarks of cancer that may be present. We developed an algorithm for detecting microcalcification clusters, the most common and useful signs of early, potentially curable breast cancer. We describe this algorithm, which utilizes contour map representations of digitized mammographic films, and discuss its benefits in overcoming difficulties often encountered in algorithmic approaches to radiographic image processing. We present experimental analyses of mammographic films employing this contour-based algorithm and discuss practical issues relevant to its use in an automated film interpretation instrument.

Noise reduction in biological step signals: application to saccadic EOG

A weighted filter for noise reduction in nonrecurrent step signals where adaptive filtering cannot be applied is described. An optimal correction of a conventional finite impulse response (FIR) filter is achieved by using a priori knowledge of noise variance and a continuous estimation of the error signals power. The weighted filter provides an optimal compromise between noise filtering and distortionless tracking. The prior knowledge required is that of the noise power and the lowest frequency in the noise spectrum. Application of the weighted filter to the saccadic electro-oculogram (EOG) results in better estimations of saccade duration and velocity.

Detection of microcalcification clusters using neural networks

One of the earliest mammographic signs of breast cancer, a cluster of microcalcifications, is difficult to detect visually, due to the small size of microcalcifications and their resemblance to other bright structures in mammograms. A fully automated algorithm that we developed for detecting clusters of microcalcifications extracts features that represent individual microstructures using the contour map of the mammogram. This allows computations without using predetermined areas of interest (kernels). The extracted features quantify visual recognition criteria. Microcalcifications are discriminated from other microstructures using multi-layer feedforward neural networks whose inputs are the extracted features.<<ETX>>

An algorithm for early breast cancer detection in mammograms

Presents a novel algorithm specifically designed for detecting clusters of microcalcifications that are early mammographic signs of breast cancer. The algorithm, which can be implemented in a general-purpose computer, is intended to assist the radiologist by indicating the location of suspicious clusters. A small number of parameters (features) are extracted from the mammogram and used in a decision-making process that requires no human supervision.<<ETX>>