An easily implemented method for abbreviation expansion for the medical domain in Japanese text. A preliminary study.

BACKGROUND: One of the barriers for the effective use of computerized health-care related text is the ambiguity of abbreviations. To date, the task of disambiguating abbreviations has been treated as a classification task based on surrounding words. Application of this framework for languages that have no word boundaries requires pre-processing to segment a sentence into separate word sequences. While the segmentation processing is often a source of problem, it is unknown whether word information is really requisite for abbreviation expansion. OBJECTIVES: The present study examined and compared abbreviation expansion methods with and without the incorporation of word information as a preliminary study. METHODS: We implemented two abbreviation expansion methods: 1) a morpheme-based method that relied on word information and therefore required pre-processing, and 2) a character-based method that relied on simple character information. We compared the expansion accuracies for these two methods using eight medical abbreviations. Experimental data were automatically built as a pseudo-annotated corpus using the Internet. RESULTS: As a result of the experiment, accuracies for the character-based method were from 0.890 to 0.942 while accuracies for the morpheme-based method were from 0.796 to 0.932. The character-based method significantly outperformed the morpheme-based method for three of the eight abbreviations (p < 0.05). For the remaining five abbreviations, no significant differences were found between the two methods. CONCLUSIONS: Character information may be a good alternative in terms of simplicity to morphological information for abbreviation expansion in English medical abbreviations appeared in Japanese texts on the Internet.

Cryogenic trapping for determination of odor concentration.

The sensory testing method applied under Japanese law to measure odor concentration has a lower detection limit of 10 in the specified Odor Index. To measure odor below the limit, a condensing procedure using solid sorbents (Tenax-TA, Unicarbon B and Carbosieve SIII) has been developed and used in Japan. This procedure however cannot condense all odorous substances, and is specifically unsuited to hydrogen sulfide, methyl mercaptan, dimethyl sulfide, ammonia, and other typical odorous substances. In the present study, cryogenic trapping was tested to improve recovery rate. As water in sample air causes choking of the trap tube, vacant pre-columns to condense the water were connected to the Tenax-TA-packed column. The columns were chilled with liquid oxygen before passage of 100 L of sample air. The columns were then heated to 200 degrees C under passage of 50 mL/min of nitrogen carrier gas to desorb odors. The desorbed gases were captured in sampling bags made of polyethylene terephthalate film. The total volume of desorbed gases was approximately 1 L. The method showed good recovery rates for hydrogen sulfide, methyl mercaptan, dimethyl sulfide and ammonia, and was useful for determining low-level odor concentrations during measurement of odor in ambient air at various sites in Osaka City.

Spatiotemporal patterns of movement-related fields in stroke patients.

In this study, brain functions in stroke patients and normal subjects were analyzed by observing neuromagnetic fields during button pressing tasks. The measurements included force measurement, visual stimuli presentation and magnetoencephalography (MEG). A 122-channel whole-head MEG system (Neuromag 122) was used. A total of 18 subjects (11 post-stroke and 7 control subjects) participated in the study; adequate MEG data were obtained from 3 of the patients. Equivalent current dipoles for MRFs were estimated with the following parameters: goodness of fit (> 75%), confidence volume (< 6000 mm3), and the duration of dipole (> 10 msec). In addition to the single dipole analysis, Minimum Current Estimates were applied for source estimation since neural activities for stroke patients were observed at the motor cortex of the contralateral side as well as other areas of the brain. Contralateral motor cortex was activated for the normal subjects at 50 ms prior to the force onset, whereas ipsilateral motor cortex was activated for some stroke patients and patients' dipole moments differed not only in their locations but also in their latencies, ranging from -150 to 50 ms to the force onset. The results were in agreement with the findings by PET and fMRI studies; therefore, it was implied that the compensating motor functions were shifted to neighboring areas of the brain due to the recovering motor function after stroke.

Rejection of stimulus-related MEG artifacts using independent component analysis.

Electro-gustatory (EG) stimuli have been used to investigate the gustatory information processing in the human brain with precise temporal and spatial accuracy. But this technique was not widely applicable to magnetoencephalographic measurements due to the difficulty in eliminating huge stimulus-related artifact. In this study, we used independent component decomposition of the measured signal to reject the stimulus-related artifact from other brain activities based on information maximization (infomax) approach. Infomax ICA was applied to raw MEG data measured by 122 magnetometer channels producing 122 temporally independent components, and the trials for each component were averaged separately. Each independent component was visually inspected and the components obviously representing the EG stimulus-related artifacts were excluded from remixing and the signal projection onto the original MEG signal space to reconstruct the artifact-free MEG signals. Results showed that large EG stimulus-related artifacts were clearly removed from MEG waveforms. We were also able to separate blink-related components and magnetocardiographic components as well as the components representing alpha-band (8-13 Hz) spontaneous brain activity.

MEG measurement of event-related brain activity evoked by emotional prosody recognition.

Cortical areas involved in processing of emotional prosody (EP) in spoken language, such as joy or sadness, have been found in functional magnetic resonance imaging (fMRI) studies bilaterally or dominantly in the right frontal or temporal lobes. In this study, we investigated spatiotemporal patterns of cortical activity related to EP processing using magnetoencephalography (MEG). In this experiment, a joyful face (JF) or a sad face (SF) was displayed after voices which had emotional features of joy (joy prosody: JP) or sadness (sad prosody: SP) were presented. Subjects were requested to judge whether emotional features of the voice and the face were identical or not. MEG signals evoked by emotional voices were measured and significant differences of cortical activities associated with processing of emotional feature were observed between the right and left hemisphere during the latency of 100-150 ms that includes the N1m component. Our study suggests that MEG is a useful method, in addition to fMRI and event-related scalp potentials (ERP) for studying non-invasively EP processing in the human brain.

Effect of word familiarity on visually evoked magnetic fields.

This study investigated the effect of word familiarity of visual stimuli on the word recognizing function of the human brain. Word familiarity is an index of the relative ease of word perception, and is characterized by facilitation and accuracy on word recognition. We studied the effect of word familiarity, using "Hiragana" (phonetic characters in Japanese orthography) characters as visual stimuli, on the elicitation of visually evoked magnetic fields with a word-naming task. The words were selected from a database of lexical properties of Japanese. The four "Hiragana" characters used were grouped and presented in 4 classes of degree of familiarity. The three components were observed in averaged waveforms of the root mean square (RMS) value on latencies at about 100 ms, 150 ms and 220 ms. The RMS value of the 220 ms component showed a significant positive correlation (F=(3/36); 5.501; p=0.035) with the value of familiarity. ECDs of the 220 ms component were observed in the intraparietal sulcus (IPS). Increments in the RMS value of the 220 ms component, which might reflect ideographical word recognition, retrieving "as a whole" were enhanced with increments of the value of familiarity. The interaction of characters, which increased with the value of familiarity, might function "as a large symbol"; and enhance a "pop-out" function with an escaping character inhibiting other characters and enhancing the segmentation of the character (as a figure) from the ground.

A new dynamical approach to auditory evoked magnetic field by blind identification.

A new approach to understand neural dynamics underlying the generation of auditory evoked magnetic field is proposed. MEG time series data are temporally decorrelated by using a blind signal separation method. Two components are selected from their periodical property and a remixing matrix is applied to the two selected components to retrieve MEG signals of auditory evoked magnetic field. After principal component data for each sensor pairs are calculated, a minimum phase innovation model is identified from the viewpoint of statistical inverse problem. By using a blind identification method based on feedback system theory transfer functions can be evaluated to get a dynamical understanding of brain auditory functions. It is reported that all changes of their impulse responses between right and left hemisphere decay within about 40 ms, and that directional differences in transfer functions can be found.

A grid computing infrastructure for MEG data analysis.

Magnetoencephalography (MEG) is widely used for studying brain functions, but clinical applications of MEG have been less prevalent. One reason is that only clinicians who have highly specialized knowledge can use MEG diagnostically, and such clinicians are found at only a few major hospitals. Another reason is that MEG data analysis is getting more and more complicated, and deals with a large amount of data, and thus requires high-performance computing. These problems can be solved by the collaboration of human and computing resources distributed in multiple facilities. A new computing infrastructure for brain scientists and clinicians in distant locations was therefore developed by the Grid technology, which provides virtual computing environments composed of geographically distributed computers and experimental devices. A prototype system connecting an MEG system at the AIST in Japan, a Grid environment composed of PC clusters at Osaka University in Japan and Nanyang Technological University in Singapore, and user terminals in Baltimore was developed. MEG data measured at the AIST were transferred in real-time through a 1-GB/s network to the PC clusters for processing by a wavelet cross-correlation method, and then monitored in Baltimore. The current system is the basic model for remote-access to MEG equipment and high-speed processing of MEG data.

Information processing of visually-induced apparent self motion in the cortex of humans: analysis with magnetoencephalography.

The cortical site which processes information on whole-body linear displacement is unknown. In this study, neuromagnetic responses to a visually-induced linear vection were recorded in 5 healthy, right-handed, adult subjects using a 122-channel whole cortex neuromagnetometer. We presented expanding rectangles on the screen which came into view one after another and accelerated in expanding speed at random cycle, giving the subjects the sensation of linear self motion (linear vection) through an illusory tunnel with occasional acceleration. Clear responses of magnetic fields related to the accelerative event were obtained in both hemispheres around the parietal and temporal regions. The dipole sources of the component were estimated in the cortex around the superior temporal sulcus, insula and medial superior temporal area. Some parts of these regions may have been comprised in the vestibular cortex, suggesting that it processes the sensation of linear self motion and plays an important role in space perception.

Frontal midline theta rhythms reflect alternative activation of prefrontal cortex and anterior cingulate cortex in humans.

Frontal midline theta rhythm (Fm theta) often appears on electroencephalogram (EEG) during consecutive mental tasks. To clarify the source of rhythmic activity, magnetoencephalogram (MEG) and EEG were simultaneously measured in six healthy volunteers during different mental tasks using whole head MEG system. MEG records were averaged every one cycle of Fm theta rhythms using individual positive peaks of Fm theta waves in Fz EEG as a trigger. Averaged theta components of MEG signals were analyzed with a multi-dipole model. Two sources were estimated to the regions both of the prefrontal-medial superficial cortex and anterior cingulate cortex (ACC). These regions were alternatively activated in about 40 to 120 degrees phase shift during one Fm theta cycle. From above results, we hypothesize that appearance of Fm theta during consecutive mental tasks reflects alternative activities of the medial prefrontal cortex and ACC.

A pitch glide activates an intermediate response between auditory N1 and mismatch negativity.

Pitch glides of a continuous tone elicit auditory N1-like responses. However, their characteristics have not well been investigated, and it remained unclear whether the response is an auditory true N1 or the mismatch negativity (MMN). We found here that a rapid pitch glide activates almost the same response as a true N1. On the contrary, as the rate of the pitch glide decreases, the response continuously varies the characteristics from true N1 to MMN. This suggests that there would exist intermediate responses between auditory N1 and MMN.

Dynamic cortical activation in mental image processing revealed by biomagnetic measurement.

The mental rotation task has been reported to activate the human parietal and extra-striate areas, based on the results of fMRI and PET analysis. In the present study, we investigated the dynamic properties of the distributed cortical activity related to mental rotation processes at high temporal resolution by means of brain magnetic field measurements and a linear inversion algorithm. Distributed neural activities during the mental rotation and control tasks were estimated for six subjects, and the differences in the activity distribution were analyzed. Statistically significant differences in the parietal and lateral posterior temporal region were detected 200-300 ms after the visual stimulus, indicating that the dorsal and ventral pathway were included in the mental image processing.

Interhemispheric asymmetry of event-related fields concerned with logical processing during auditory oddball stimulation.

This study investigated the cortex during logical processing of auditory information using a 122-channel dc-SQUID gradiometer. The experimental task was designed to require a simple logical decision prior to counting rare paired tones, which consisted of two different pitches and were presented to separate ears. Among six subjects, left and right predominant dipolar activity was observed in three subjects each. When the dipolar sources were superimposed on MR images, the inferior region of the supramarginal gyrus showed activation, suggesting that logical processing occurred in the association cortex but not in the auditory cortex. We propose a modified cognitive sequence model in which auditory information processed in Heschl's gyri is transmitted to the supramarginal gyrus to commence automatic detection processing.

Auditory evoked off-response: its source distribution is different from that of on-response.

Offset auditory responses were investigated by electroencephalography mainly in the 1970s, but since then no particular attention has been paid to them. Among the studies using magnetoencephalography (MEG) devices there are, to our knowledge, only three studies of the auditory off-response, and no significant variance has ever been observed between the source locations of on- and off-responses elicited from pure tones. We measured auditory evoked magnetic fields (AEFs) to various frequency pure tone stimulation in 5 healthy subjects with a 122-channel helmet-shaped magnetometer, and compared the distributions of the source locations of auditory N100m-Off (magnetic off-response around 100 ms) with those of N100m-On. Their spatial distributions were quite close to each other, and yet they were significantly different.

Neuromagnetic activity in the human left cerebral hemisphere concerning logical processing during auditory oddball stimulation.

The aim of this study was to examine the cortex during the logical processing of auditory information using a whole-cortex type dc-SQUID gradiometer. The task modified the normal auditory oddball paradigm to require the processing of simple logic before counting a rare stimulus mentally. Although the latency of P300 m did not change, a dipolar magnetic field pattern was observed over the left cerebral hemisphere at approximately 280 ms poststimulus before forming the field pattern of P300 m. The equivalent current dipole source was estimated to be medial to the N100 m source. It was suggested that the additional load of logic processing may activate Heschl's gyri in the left hemisphere.

Task-dependent laterality for cue decoding during spoken language processing.

The task-dependent laterality of the auditory cortices was investigated by measuring the magnetic fields elicited by three forms of a Japanese verb, which differed in terms of prosodic and phonetic cues. Significant task-dependent magnetic fields were found in both hemispheres during a prosody-related task, but only in the left during a phoneme-related task. The latency was similar to the mismatch negatively which reflects the neural activity of automatic cue decoding. These results suggest that task-dependent schemata are activated at least partially in parallel with automatic cue-decoding processes such that those in the left hemisphere process linguistic information irrespective of acoustic cues whereas those in the right hemisphere process prosodic information.

Ipsilateral dominance of human olfactory activated centers estimated from event-related magnetic fields measured by 122-channel whole-head neuromagnetometer using odorant stimuli synchronized with respirations.

The aim of this study was to measure and analyze olfactory event-related magnetic fields using a whole-cortex biomagnetometer (122-channel SQUID gradiometer). Amyl-acetate gas (approx. 1%) was administered for 300 msec into either the right or left nostril in synchronization with respiration using a mask and an optical fiber sensor. Clear olfactory event-related magnetic fields were asymmetrically obtained on both sides of the forehead in all six subjects. The generators of olfactory magnetic fields were estimated at two regions located fairly asymmetrivally near the bilateral frontal deep areas. The goodness-of-fit was better for the two-dipole model than the one-dipole model in all experiments. In almost all subjects the latency and intensity of ipsilateral olfactory magnetoencephalography (MEG) responses were shorter and larger than those of the contralateral responses, respectively. These results suggest that the olfactory MEG responses on the ipsilateral side are generally larger and more dominant than those on the contralateral side in the human olfactory system.

Tuning specificities to aliphatic odorants in mouse olfactory receptor neurons and their local distribution.

1. Odor responses to two homologous series of n-fatty acids (nFA) and n-aliphatic alcohols (nAA) with a straight chain of three to nine carbons were examined by measuring odor-induced [Ca2+]i increase in mouse olfactory receptor neurons (ORNs) isolated by the tissue-printing method. 2. One-third of the ORNs responsive to nFA and/or nAA were alternately sensitive to either type of odorant. Their sensitivities were usually near maximal for one or two odorants and decreased with differences in the carbon chain length from the tuned odorants. 3. Two-thirds of the ORNs responsive to nFA and/or nAA were sensitive to both types of odorants. Most of them were also tuned to one or two odorants in each series with similar carbon chain lengths and showed a decrease of sensitivity with increasing stereochemical discrepancy, similar to nFA/nAA discriminating ORNs. 4. In 10 of 20 non-nFA/nAA discriminating ORNs, the sensitivity to nFA was > 10 times greater than to nAA, and 80% of them were localized in a central region of olfactory epithelium on the septum wall where ORNs preferentially project to the dorsomedial or centromedial regions of the olfactory bulb. In addition, the sensitivity to three series of n-aliphatic odorants with an added amino group was examined. Sensitivity became higher as the electronegativity of the functional groups increased, suggesting that a hydrogen bond might partly mediate affinity in one type of non-nFA/nAA discriminating ORNs. 5. The diversity in odorant tuning specificity and sensitivity of the individual ORNs indicated that their receptor sites were finely tuned to the stereochemical structures of numerous odorants by changes in the three-dimensional size and intermolecular positions of the hydrophobic domains for hydrophobic bond, as well as the proton-acceptor or donor for the hydrogen bond and the electrical charge for the ionic bond. 6. The subpopulation of ORNs tuned to an individual odorant increased as the length of carbon chain of the odorant increased from three to nine. This tendency was more marked for nFA than for nAA in the case of non-nFA/nAA discriminating ORNs. 7. Data obtained by the in vitro approach using the tissue-printing method suggested that three or more subtypes of ORNs, which were similar in some cases and significantly different in other cases, were located within close proximity to one another.