A new method (BIM) by bioelectrical impedance for the estimation of human whole body or local part fat volume is described. This method is based upon that the resistivity of adipose tissue is much greater than that of fat-free mass of which frequency characteristics shows β dispersion phenomenon. It is well known that the admittance locus of fat-free mass provides the semicircle. Because the impedance value of trunk is only 10 or 20% of the whole body impedance (Z_all) between arm and leg used by the commercialized impedance fat meter, if the fat volume in the trunk increase, the estimation error in the fat volume estimated from Z_all would increase. Therefore, instead of Z_all, BIM adopts the impedances measured at 5 local parts which are antebrachium, brachium, trunk, femoral part and cruralis. We measured the impedance of each parts for the frequency range from 3kHz to 300kHz and calculate the admittance locus in order to obtain the conductance at much lower frequency (G_e) and much higher frequency (G_inf) than β dispersion frequency. Arms, trunk and legs are represented by parallel models which are cone shape and consists of fat layer and fat-free mass layer. And then we calculate the equivalent conductivity σ of parallel models from G_e or G_inf. From σ, fat volumes of local parts which are arm, trunk and leg and whole body are estimated. When we estimate the fat volume of trunk, in order to take into consideration the adipose tissue which distributes around organs in trunk like liver, stomach, and etc., the weighting function K which is statistically given is multiplied to the equivalent conductivity of trunk. The reference fat volume can be measured by hydrometry. We can conclude that (1) the impedance at the trunk must be measured to consider the fat volume in the trunk, (2) G_e or G_inf must be used instead of impedance at about 50kHz to estimate the fat volume, and (3) correlation coefficients between the whole fat volume measured by hydrometry and that by BIM with G_e and G_inf are about 0.77 and 0.94, respectively.

View full abstract

We have developed a method of measuring the three components of eye movement noninvasively by using an image analyzing system. The horizontal and vertical is measured to track the pupil and the torsional is measured to track a particular iris striation automatically. The system consists of a high resolution infrared CCD camera system with a pair of goggles and an eye movement analysis system using computerized image processing. A new pulse eye illumination method of the high resolution camera system makes it possible to take clear images of the pupil and iris striations, so the images can be analyzed automatically and continuously. We report the performance of our method as estimated with a simulation system, the technique of our system in detail and the clinical data of three components of the positioning nystagmus, etc. This system could be used for clinical examination of vertigo.

View full abstract

We often encounter the contamination of EEG with various artifacts in spite of a technician's effort to eliminate those artifacts during the recording. The electroencephalographer (EEGer) always pays great attention to discriminate the artifacts from the EEG signals, in order to avoid misinterpretation of the record due to the artifacts. However, this problem has not been resolved in the automatic EEG interpretation system which we have developed previously. This paper describes a method for automatic detection of artifacts in EEG records. Artifacts treated in this paper contained blink, horizontal eye movement, artifact originating from ear-lobe reference electrode and EMG artifacts. The awake EEG records obtained from 33 patients were used to develop the method and verify of its performance. The equations for detecting the artifacts were determined by considering the respective features and taking into account the result of visual inspection of artifacts done by an EEGer. Results of verification revealed that the accuracy of the proposed automatic detection of artifacts was good enough for pre-processing of automatic EEG interpretation, and the automatic EEG interpretation employing the present method broadened the range of application even in the record contaminated with artifacts.

View full abstract

This paper presents an automatic extraction system (called TOPS-3D: Top Down Parallel Pattern Recognition System for 3D Images) of soft tissues from 3D MRI head images by using model driven analysis algorithm. As the construction of system TOPS we developed, two concepts have been considered in the design of system TOPS-3D. One is the system has a hierarchical structure of reasoning using model information in higher level, and the other is a parallel image processing structure is used to extract plural candidate regions for a destination entity. The new points of system TOPS-3D are as follows. (1) The TOPS-3D is a three-dimensional image analysis system including 3D model construction and 3D image processing techniques. (2) A technique is proposed to increase connectivity between knowledge processing in higher level and image processing in lower level. The technique is realized by applying opening operation of mathematical morphology, in which a structural model function defined in higher level by knowledge representation is immediately used to the filter function of opening operation as image processing in lower level. The system TOPS-3D applied to 3D MRI head images consists of three levels. First and second level are reasoning part, and third level is image processing part. In experiments, we applied 5 samples of 3D MRI head images with size 128×128×128 voxels to the system TOPS-3D to extract the regions of soft tissues such as cerebrum, cerebellum and brain stem. From the experimental results, the system is robust for variation of input data by using model information, and the position and shape of soft tissues are extracted corresponding to anatomical structure.

View full abstract

It is said that the presence of spicula is the most important of all visual signs for which radiologists search, to determine cancer malignancy. In this paper, we propose a new method to automatically perform this spicula recognition/classification process. This new method, first enhances ridges (which can be considered to be spicula shadows) using a newly developed process called Tophat by Partial Recognition. Next, bar filter opening is applied to remove noise components left from the previous process. Finally, spicula concentration (line concentration) is computed using a graylevel extension of the conventional line concentration equation. This method was tested on 24 samples with 7 containing confirmed spicula cancers. The method showed a 100% recognition/classification rate for spicula cancers with 0% false positives. The recognition/classification rates stated, although on a limited number of images, are very promising. Comparison to other published works is not possible without a common database, however the results showed certainly seem competitive.

View full abstract

The conventional automatic blood cell classifier uses a set of statistical discriminant functions composed of multi stage, tree structure, linear and second order functions. In this paper, we tried to replace these discriminant functions with three layer neural networks (NN). In this case, chosing the number of hidden layer units becomes a problem. So, we focus our discussion mainly on the optimization of the number of hidden layer units, using a gradual reduction method (GRM), a method we previously proposed. In this paper, we use GRM and examined its effectiveness. We used and compared two types of back propagation (BP) algorithms, standard BP (Std-BP) and BP with forgetting (BP-F), with GRM. From the results, GRM+BP-F is found to suit the above purpose better, from the stability of the recognition curve and from its independency to initial weight conditions. Further the recognition rate is stable, i. e., we don't need to learn NN again. Also we found that the computation time can be shortened with GRM+BP-F as compared to the case of learning using repeated BP alone.

View full abstract

Articulation of words in tinnitus patients following electrical stimulation of the cochlea was assessed by means of a grammatically correct but non-sense Japanese 4-segment sentence articulation test. Improved articulation of words was reported in 19 out of 25 patients following electrical stimulation of the cochlea. In 15 out of 19 patients, the incorrect number of grammatically correct but non-sense Japanese 4-segment sentences significantly decreased, showing that electrical stimulation of the cochlea improved articulation of words. In 8 out of 15 patients, improved articulation of words was observed in the ear contralateral to the ear treated, suggesting that improved articulation of words, at least, in these patients occurred in the central auditory system. These results suggested that electrical stimulation of the cochlea improved articulation of words in tinnitus patients when tinnitus decreased. Based on these results, we proposed the concept of functinal electrical stimulation in the auditory nervous system.

View full abstract

The influence of intensity of coronary arterial stenosis on intramyocardial blood flow was examined by a model study. We composed a myocardial circulation model incorporating the intramyocardial pressure pump and the time varing elastance as the extravascular compressive force. The model has two myocardial layers, epimyocardial and endomyocardial layers, which consist of arteriolar, capillary and venular compartments, respectively. Stenosis was introduced into the epicardial coronary artery, and the pressure loss across stenosis was expressed by the sum of friction loss and separation loss. With increment of stenotic severity, we compared the arteriolar blood pressure and blood flow waveforms during systole and diastole between the epimyocardium and the endomyocardium. Following the simulation study, it was reveald that in the endomyocardium, the increase of systolic reverse flow and the decrease of diastolic forward flow were concomitantly progressd with the severity of stenosis, leading to remarkable reduction of total volume flow. On the other hand, despite the decrease of diastolic forward flow, systolic reverse flow did not develop in the epimyocardium until stenosis was severe enough to cause extreme pressure drop. Thus, epimyocardial ischemia was not apparent until stenosis became very severe, i. e., about 97% area stenosis. In coclusion, the endomyocardium is susceptible to ischemia due to the reduction of diastolic forward flow with the concomitant increment of systolic reverse flow, whereas the epimyocardium is tolerant to ischemia, resulted from the relative lack of systolic reverse flow disturbance along the progression of stenosis.

View full abstract

In evaluating histological images obtained from pathological samples, taking advantage of the fact that each tissue component exhibits specific staining characteristics. We propose a method for classification of hematoxylin and eosin (HE)-stained breast tumor tissues as benign or malignant type. Two types of neural networks are used in the two-step process consisting of the extraction of features from HE staining images and evaluation of their values. In the feature extraction stage, regions in HE staining images are divided into four categories, nucleus, cytoplasm, interstitium and background, on the basis of color information. In the evaluation stage, values of the extracted features based on stainability (percentage area of nucleus, cytoplasm and interstitium, frequency distribution and size) are used. The method developed was applied to 30 test images from 17 cases consisting of 10 different types of tissues and to 26 images from 13 cases of blind samples. The accuracy of classification of breast tumors as benign or malignant was excellent, and the effectiveness of the method was confirmed. The proposed method has two advantages: (1) it is effective in the evaluation of staining images that exhibit considerable variability such as in contrast (variation in input images) and (2) applications to other types of HE-stained tumors are possible by the use of the algorithm proposed in this study.

View full abstract

An ideal reference for Unipolar electrocardiography is the electrode at infinity without any potential variations. Wilson's electrode was devised to approximate this ideal zero reference. Being constructed only from both arms and left leg, his device is quite practical and convenient. However, better approximation is conceivable with a connection of many electrodes on the body surface. In this communication, we propose two methods for the calculation of an ideal reference potential from potential distributions measured with any reference system. One of the methods was applied to certain clinical cases to confirm its consistency and to evalute the potential variation of Wilson's central terminal. In the first method, the potential φ_out for the space outside the body is determined such that it coincides with the measured potential distribution on the body surface within an unknown constant. This constant is then determined from the behavior of φ_out at infinity. The second method is based on the immersion of the body into an infinite homogeneous meium with the same conductivity as that of the body. Then, the outside conductivity is made to approach zero. The second method was applied to 60 clinical cases including normal persons and cases with myocardial infarction, bundle branch blocks and WPW syndrome. The results indicated that the maximal potential of Wilson's terminal was in the range of -0.4 to 0.2mV, which amounted to -10 to 5% of the surface voltage in most of the cases. These values are consistent with previous reports on the bath immersion of body and the others. In conclusion, this communication answers the fundamental question of unipolar electrocardiography on the sound theoretical bases.

View full abstract

Event-related potentials (ERPs) were recorded from seven subjects, using twelve scalp electrodes and three other electrodes, monitoring eye and throat movements. The experiments were divided into two blocks. In one block, Japanese katakana letters (phonograms) or unreadable pseudo letters, that were made by deforming letters, were visually presented at random. The subject was instructed to silently read when readable letters were presented and to withhold it when pseudo letters were presented (control state). Fifty letters and fifty pseudo letters were presented, one at a time in random order. In the other block, meaningless words having three katakana letters, or words having three pseudo letters were presented. Each subject carried out three or four of each block. The average ERPs with silent reading minus those without silent reading showed five peaks: a negative peak at a 400-ms latency in the left posterior temporal scalp area, a positive peak at a 400-ms latency in the frontal scalp area, a positive peak at a 520-ms latency in the parietal scalp area, a negative peak at a 620-ms latency in the frontal, parietal, and temporal scalp area, and a negative peak at a 700-ms latency in the frontal scalp area. These peaks, except for the 520-ms positive peak, were larger for words than for letters. The present silent-reading task included mental processes involving recognition of letters, judgement of whether the stimulus is for silent speech, and silent speech itself, but the control task did not. The negative difference in the left posterior temporal scalp area was not observed in the previous ERP experiments using silent speech prompted by light-emitting diodes, and is considered as related to recognize the presented letters. The positive difference in the parietal scalp area was observed in the previous experiments, and is considered as related to the P3 component for target detection. The frontal positive and negative differences were also observed in the previous ERP experiments, and are considered to be related to silent speech.

View full abstract

Magnetization transfer contrast (MTC) using fast spin echo (FSE) was evaluated with phantom and human breast. Phantom consists of oil, 0.9% NaCl and gelatin with various gelatin-water ratio. On multi slice FSE images, the attenuation of signal intensity in gelatin and mammary gland due to MTC was observed. Magnetization transfer ratio (MTR) was calculated from the center slice acquired with multi slice FSE and the image acquired with single slice FSE. Phantom examination proved the accuracy of MTC using FSE to be valid. MTR of fat tissue was less than that of mammary gland. MTR of carcinoma was more than that of mammary gland. This study suggests that MTR might be a new indicator for assessment of tumor tissue characterization.

View full abstract