Using an artificial neural network to diagnose hepatic masses

Using abdominal ultrasonographic data and laboratory tests, radiologists often find differential diagnoses of hepatic masses difficult. A computerized second opinion would be especially helpful for clinicians in diagnosing liver cancer because of the difficulty of such diagnoses. A back-propagation neural network was designed to diagnose five classifications of hepatic masses: hepatoma, metastic carcinoma, abscess, cavernous hemangioma, and cirrhosis. The network input consisted of 35 numbers per patient case that represented ultrasonographic data and laboratory tests. The network architecture had 35 elements in the input layer, two hidden layers of 35 elements each, and 5 elements in the output layer. After being trained to a learning tolerance of 1%, the network classified hepatic masses correctly in 48 of 64 cases. An accuracy of 75% is higher than the 50% scored by the average radiology resident in training but lower than the 90% scored by the typical board-certified radiologist. When sufficiently sophisticated, a neural network may significantly improve the analysis of hepatic-mass radiographs.     

Using neural networks to diagnose cancer

While artificial brains are in the realm of science fiction, artificial neural networks (ANNs) are scientific facts. An artificial neural network is a computational structure modeled somewhat on the neural structure of the brain; both have many highly interconnected processing elements. These biologically inspired processing elements are taught by feeding examples until the results are acceptable. In the past 5 years, neural networks have become successful in providing meaningful second opinions in clinical diagnosis. In our research, a prototype artificial neural network was trained on numeral ultrasound data of 52 actual cases and then correctly identified renal cell carcinoma from renal cysts and other conditions without diagnostic errors. Our nonlinear artificial neural network was trained on software using the standard backpropagation paradigm on a 80386 microcomputer. Our ANN learned from ultrasound data in 52 cases (17 malignant, 30 cysts, and 5 other) at a Memphis hospital. The trained prototype performed without error on 47 cases which were not in the data used for training. This prototype must be validated by extending this study to more cases.     

Optical imaging of temporal integration in human auditory cortex

Behavioral and physiological studies have indicated the existence of a temporal window of auditory integration (TWI), within which similar sounds are perceptually grouped. The current study exploits the combined temporal and spatial resolution of fast optical imaging (the event-related optical signal, EROS) to show that brain activity elicited by sounds within and outside the TWI differs in location and latency. In a previous event-related brain potential (ERP) study [Sable, Gratton, and Fabiani (2003) European Journal of Neuroscience, 17, 2492-2496], we found that the mismatch negativity (MMN; a brain response to acoustic irregularities) elicited by deviations in stimulus onset asynchronies (SOAs) had a unique shape when the deviant SOA was within the TWI. In the present study, we extended these ERP results using EROS. Participants heard trains of five tones. The first four tones had SOAs of 96, 192, 288 or 384 ms. The SOA of the fourth and fifth tones was either the same (standard) or one of the other three (deviant) SOAs. With a deviant SOA of 96 ms, the cortical response was approximately 2 cm anterior to responses to longer SOA deviants, and was followed by a later response that was absent in the other conditions. Similarly to the electrical MMN, the optical mismatch response amplitudes were proportional to the magnitude of interval deviance. These results, in combination with our previous findings, indicate that the temporal integration of sounds is reflected in cortical mismatch responses that differ from the typical response to interval deviance.     

Proof‐of‐concept evidence for trimodal simultaneous investigation of human brain function

The link between spatial (where) and temporal (when) aspects of the neural correlates of most psychological phenomena is not clear. Elucidation of this relation, which is crucial to fully understand human brain function, requires integration across multiple brain imaging modalities and cognitive tasks that reliably modulate the engagement of the brain systems of interest. By overcoming the methodological challenges posed by simultaneous recordings, the present report provides proof‐of‐concept evidence for a novel approach using three complementary imaging modalities: functional magnetic resonance imaging (fMRI), event‐related potentials (ERPs), and event‐related optical signals (EROS). Using the emotional oddball task, a paradigm that taps into both cognitive and affective aspects of processing, we show the feasibility of capturing converging and complementary measures of brain function that are not currently attainable using traditional unimodal or other multimodal approaches. This opens up unprecedented possibilities to clarify spatiotemporal integration of brain function.     

Validation of a method for coregistering scalp recording locations with 3D structural MR images

A common problem in brain imaging is how to most appropriately coregister anatomical and functional data sets into a common space. For surface-based recordings such as the event related optical signal (EROS), near-infrared spectroscopy (NIRS), event-related potentials (ERPs), and magnetoencephalography (MEG), alignment is typically done using either (1) a landmark-based method involving placement of surface markers that can be detected in both modalities; or (2) surface-fitting alignment that samples many points on the surface of the head in the functional space and aligns those points to the surface of the anatomical image. Here we compare these two approaches and advocate a combination of the two in order to optimize coregistration of EROS and NIRS data with structural magnetic resonance images (sMRI). Digitized 3D sensor locations obtained with a Polhemus digitizer can be effectively coregistered with sMRI using fiducial alignment as an initial guess followed by a Marquardt-Levenberg least-squares rigid-body transform (df = 6) to match the surfaces. Additional scaling parameters (df = 3) and point-by-point surface constraints can also be employed to further improve fitting. These alignment procedures place the lower-bound residual error at 1.3 +/- 0.1 mm (micro +/- s) and the upper-bound target registration error at 4.4 +/- 0.6 mm (micro +/- s). The dependence of such errors on scalp segmentation, number of registration points, and initial guess is also investigated. By optimizing alignment techniques, anatomical localization of surface recordings can be improved in individual subjects.     

Short-Term Memory During Stage-2 Sleep

A recently-proposed model of dream recall phenomena (Koulack & Goodenough, 1976) assumes a short-term memory store which functions during sleep, and from which either dreams or externally-delivered stimuli can be retrieved if the subject is awakened during the life of the short-term trace. To test this assumption, two experiments were designed to examine short-term memory for auditory stimuli over intervals of uninterrupted stage-2 sleep. A random series of single-digit numbers was presented at the rate of one every 30 sec throughout the night, at an intensity slightly above normal conversational levels. Subjects habituated to the sound and did not awaken spontaneously. Experimental awakenings took place either at 1 sec (Experiment 1) or at 1, 5, or 10 sec (Experiment 2) after selected target stimuli, and the subjects were asked to recall which number they had heard. Significant recall was obtained over intervals of up to 10 sec, and the likelihood of recall was inversely related to reaction times to the awakening stimulus. Repeating the target numbers at 30-sec intervals increased EEG alpha activity and K-complex amplitude, suggesting that some form of trace persists for at least 30 sec.     

A naturally occurring genetic variant in the human chorionic gonadotropin-beta gene 5 is assembly inefficient.

The hCGbeta gene family is composed of six homologous genes linked in tandem repeat on chromosome 19; the order of the genes is 7, 8, 5, 1, 2, and 3. Previous studies have shown that hCGbeta gene 5 is highly expressed during the first trimester of pregnancy. The purpose of our study was to identify naturally occurring polymorphisms in hCGbeta gene 5 and determine whether these alterations affected hCG function. The data presented here show that hCGbeta gene 5 was highly conserved in the 334 asymptomatic individuals and 41 infertile patients examined for polymorphisms using PCR followed by single stranded conformational polymorphism analysis. Most of the polymorphisms detected were either silent or located in intron regions. However, one genetic variant identified in beta gene 5 exon 3 was a G to A transition that changed the naturally occurring valine residue to methionine in codon 79 (V79M) in 4.2% of the random population studied. The V79M polymorphism was always linked to a silent C to T transition in codon 82 (tyrosine). To determine whether betaV79M hCG had biological properties that differed from those of wild-type hCG, a beta-subunit containing the V79M substitution was created by site-directed mutagenesis and was coexpressed with the glycoprotein hormone alpha-subunit in Chinese hamster ovary cells and 293T cells. When we examined betaV79M hCG biosynthesis, we detected atypical betaV79M hCG folding intermediates, including a betaV79M conformational variant that resulted in a beta-subunit with impaired ability to assemble with the alpha-subunit. The inefficient assembly of betaV79M hCG appeared to be independent of beta-subunit glycosylation or of the cell type studied, but, rather, was due to the inability of the betaV79M subunit to fold correctly. The majority of the V79M beta-subunit synthesized was secreted as unassembled free beta. Although the amount of alphabeta hCG heterodimer formed and secreted by betaV79M-producing cells was less than that by wild-type beta-producing cells, the hCG that was secreted as alphabeta V79M heterodimer exhibited biological activity indistinguishable from that of wild-type hCG.     

Design and pharmacological activity of phosphinic acid based NAALADase inhibitors.

A novel series of phosphinic acid based inhibitors of the neuropeptidase NAALADase are described in this work. This series of compounds is the most potent series of inhibitors of the enzyme described to date. In addition, we have shown that these compounds are protective in animal models of neurodegeneration. Compound 34 significantly prevented neurodegeneration in a middle cerebral artery occlusion model of cerebral ischemia. In addition, in the chronic constrictive model of neuropathic pain, compound 34 significantly attenuated the hypersensitivity observed with saline-treated animals. These data suggest that NAALADase inhibition may provide a new approach for the treatment of both neurodegenerative disorders and peripheral neuropathies.     

Progesterone:estradiol ratios at implantation in ongoing pregnancies, abortions, and nonconception cycles resulting from ovulation induction

The purpose of this study was to compare progesterone (P):estradiol (E2) ratios after ovulation induction at the time of implantation in cycles resulting in ongoing pregnancies or abortions and in nonconception cycles. Material included 43 stimulated conception cycles, 29 with human menopausal gonadotropins (hMG) and human chorionic gonadotropins (hCG), 14 with clomiphene citrate (CC) with or without hCG, and 28 nonconception cycles (13 hMG and hCG, 15 CC with or without hCG). Midluteal P and E2 were measured and expressed in ng/mL. There were no differences in P:E2 ratios (mean +/- SE) for ongoing pregnancies after hMG and hCG (n = 20, 112.6 +/- 14.9), CC and hCG (n = 6, 97.0 +/- 15.9), or CC alone (n = 5, 96.2 +/- 25.5), and the data were pooled. Progesterone:estradiol ratios in 31 ongoing pregnancies and 28 nonconception cycles were 107.0 +/- 10.7 and 115.2 +/- 12.5, respectively, both significantly higher than in 12 abortions (64.5 +/- 13.2). In conclusion: (1) P:E2 ratios at the time of implantation were similar after CC with or without hCG and hMG and hCG treatment; (2) high luteal P:E2 ratio was associated with ongoing pregnancies; and (3) lower P:E2 ratio was seen in cycles leading to spontaneous abortion.     

Specificity of smooth pursuit eye movement and visual fixation abnormalities in schizophrenia. Comparison to mania and normal controls

Smooth pursuit eye movements (SPEM) were assessed in 30 schizophrenic patients, 12 lithium-free manic patients, and 20 normal controls. Compared to schizophrenic patients, manic patients evidenced less SPEM impairment in an attention enhancing, sinusoidal target motion condition and had superior performance during a visual fixation condition. SPEM and visual fixation dysfunctions may be more common in schizophrenic than in acutely manic patients, even when the latter are characterized by marked attentional dysfunction, poor interepisode psychosocial functioning, and psychosis.