Simulation studies of data classification by artificial neural networks: Potential applications in medical imaging and decision making

Artificial neural networks are being investigated in the field of medical imaging as a means to facilitate pattern recognition and patient classification. In the work reported here, the effects of internal structure and the nature of input data on the performance of neural networks were investigated systematically using computer-simulated data. Network performance was evaluated quantitatively by means of receiver operating characteristic analysis and compared with the performance of an ideal statistical decision maker. We found that the relatively simple neural networks investigated in this study can perform at the level of an ideal decision maker. These simple networks were also found to learn accurately even when the training data are extremely unbalanced with respect to the prevalence of actually positive cases and to differentiate input data patterns by recognizing their unique characteristics.     

Response of anterior parietal cortex to cutaneous flutter versus vibration.

The response of anesthetized squirrel monkey anterior parietal (SI) cortex to 25 or 200 Hz sinusoidal vertical skin displacement stimulation was studied using the method of optical intrinsic signal (OIS) imaging. Twenty-five-Hertz ("flutter") stimulation of a discrete skin site on either the hindlimb or forelimb for 3-30 s evoked a prominent increase in absorbance within cytoarchitectonic areas 3b and 1 in the contralateral hemisphere. This response was confined to those area 3b/1 regions occupied by neurons with a receptive field (RF) that includes the stimulated skin site. In contrast, same-site 200-Hz stimulation ("vibration") for 3-30 s evoked a decrease in absorbance in a much larger territory (most frequently involving areas 3b, 1, and area 3a, but in some subjects area 2 as well) than the region that undergoes an increase in absorbance during 25-Hz flutter stimulation. The increase in absorbance evoked by 25-Hz flutter developed quickly and remained relatively constant for as long as stimulation continued (stimulus duration never exceeded 30 s). At 1-3 s after stimulus onset, the response to 200-Hz stimulation, like the response to 25-Hz flutter, consisted of a localized increase in absorbance limited to the topographically appropriate region of area 3b and/or area 1. With continuing 200-Hz stimulation, however, the early response declined, and by 4-6 s after stimulus onset, it was replaced by a prominent and spatially extensive decrease in absorbance. The spike train responses of single quickly adapting (QA) neurons were recorded extracellularly during microelectrode penetrations that traverse the optically responding regions of areas 3b and 1. Onset of either 25- or 200-Hz stimulation at a site within the cutaneous RF of a QA neuron was accompanied by a substantial increase in mean spike firing rate. With continued 200-Hz stimulation, however, QA neuron mean firing rate declined rapidly (typically within 0.5-1.0 s) to a level below that recorded at the same time after onset of same-site 25-Hz stimulation. For some neurons, the mean firing rate after the initial 0.5-1 s of an exposure to 200-Hz stimulation of the RF decreased to a level below the level of background ("spontaneous") activity. The decline in both the stimulus-evoked increases in absorbance in areas 3b/1 and spike discharge activity of area 3b/1 neurons within only a few seconds of the onset of 200-Hz skin stimulation raised the possibility that the predominant effect of continuous 200-Hz stimulation for >3 s is inhibition of area 3b/1 QA neurons. This possibility was evaluated at the neuronal population level by comparing the intrinsic signal evoked in areas 3b/1 by 25-Hz skin stimulation to the intrinsic signal evoked by a same-site skin stimulus containing both 25- and 200-Hz sinusoidal components (a "complex waveform stimulus"). Such experiments revealed that the increase in absorbance evoked in areas 3b/1 by a stimulus having both 25- and 200-Hz components was substantially smaller (especially at times >3 s after stimulus onset) than the increase in absorbance evoked by "pure" 25-Hz stimulation of the same skin site. It is concluded that within a brief time (within 1-3 s) after stimulus onset, 200-Hz skin stimulation elicits a powerful inhibitory action on area 3b/1 QA neurons. The findings appear generally consistent with the suggestion that the activity of neurons in cortical regions other than areas 3b and 1 play the leading role in the processing of high-frequency (>/=200 Hz) vibrotactile stimuli.     

C1-esterase inhibitor blocks T lymphocyte proliferation and cytotoxic T lymphocyte generation in vitro

We have previously shown that activated C1s complement and activated T cells cleave beta2-microglobulin (beta2m) in vitro leading to the formation of desLys58 beta2m. This process can specifically be inhibited by C1-esterase inhibitor (C1-inh). Furthermore we showed that exogenously added desLys58 beta2m in nanomolar amounts to a one-way allogenic mixed lymphocyte culture (MLC) increased the endogenous production of IL-2 and the generation of allo-specific cytotoxic T lymphocytes. C1-inh was purified from fresh human plasma and added to human or murine MLC and mitogen-stimulated lymphocyte cultures grown in the presence of complement-inactivated serum. Read-outs were cell proliferation, lymphokine production and development of T cell-mediated cytotoxicity. We found that addition of C1-inh to MLC and mitogen- exposed murine and human lymphocyte cultures inhibited proliferation, the development of allospecific cytotoxic activity, and changed the endogenous production of IL-2, IL-4, IL-10, IL-12 and IFN-gamma. These data clearly demonstrate a regulatory function of C1-inh on T cell- mediated immune functions.      

Maximum likelihood fitting of FROC curves under an initial-detection-and-candidate-analysis model

We have developed a model for FROC curve fitting that relates the observer's FROC performance not to the ROC performance that would be obtained if the observer's responses were scored on a per image basis, but rather to a hypothesized ROC performance that the observer would obtain in the task of classifying a set of "candidate detections" as positive or negative. We adopt the assumptions of the Bunch FROC model, namely that the observer's detections are all mutually independent, as well as assumptions qualitatively similar to, but different in nature from, those made by Chakraborty in his AFROC scoring methodology. Under the assumptions of our model, we show that the observer's FROC performance is a linearly scaled version of the candidate analysis ROC curve, where the scaling factors are just given by the FROC operating point coordinates for detecting initial candidates. Further, we show that the likelihood function of the model parameters given observational data takes on a simple form, and we develop a maximum likelihood method for fitting a FROC curve to this data. FROC and AFROC curves are produced for computer vision observer datasets and compared with the results of the AFROC scoring method. Although developed primarily with computer vision schemes in mind, we hope that the methodology presented here will prove worthy of further study in other applications as well.     

How to use Chlamydia antibody testing in subfertility patients

Screening for tubal factor subfertility by means of Chlamydia antibody testing (CAT) was introduced into the initial work-up of subfertile couples several years ago. The results reported, however, are heterogeneous, and no uniformity exists in cut-off levels of titres, or in definitions of tubal factor subfertility. We performed a prospective cohort study to evaluate the implications of varying the definitions of tubal pathology and of modifying the cut-off levels on the clinical impact of CAT in predicting tubal factor subfertility. In 227 consecutive patients who attended our fertility clinic, the Chlamydia IgG antibody titre was determined and related to tuboperitoneal abnormalities at laparoscopy as a reference standard. According to received operating characteristic (ROC) curve analysis, a titre of 16 is the optimum cut-off level. Increasing the cut-off level improves specificity and positive likelihood ratio (LR+), at the expense of sensitivity and negative LR (LR-). Changing the definition of tubal factor subfertility from unspecified tuboperitoneal abnormalities into extensive adhesions and/or bilateral distal tubal occlusion improves LR+, LR- and kappa significantly. We conclude that CAT is more accurate in predicting severe distal tubal pathology than unspecified tuboperitoneal abnormalities. Although from a statistical point of view a titre of 16 is the optimum cut-off level, from a clinical point of view 32 or 64 may be preferable, depending on the aim of screening and the inception cohort.      

Circadian clock and cell cycle gene expression in mouse mammary epithelial cells and in the developing mouse mammary gland.

Mouse mammary epithelial cells (HC-11) and mammary tissues were analyzed for developmental changes in circadian clock, cellular proliferation, and differentiation marker genes. Expression of the clock genes Per1 and Bmal1 were elevated in differentiated HC-11 cells, whereas Per2 mRNA levels were higher in undifferentiated cells. This differentiation-dependent profile of clock gene expression was consistent with that observed in mouse mammary glands, as Per1 and Bmal1 mRNA levels were elevated in late pregnant and lactating mammary tissues, whereas Per2 expression was higher in proliferating virgin and early pregnant glands. In both HC-11 cells and mammary glands, elevated Per2 expression was positively correlated with c-Myc and Cyclin D1 mRNA levels, whereas Per1 and Bmal1 expression changed in conjunction with beta-casein mRNA levels. Interestingly, developmental stage had differential effects on rhythms of clock gene expression in the mammary gland. These data suggest that circadian clock genes may play a role in mouse mammary gland development and differentiation.