• Journal of Intelligence and Information Systems
• /
• v.27 no.3
• /
• pp.57-73
• /
• 2021

Maintenance and prevention of failure through anomaly detection of ICT infrastructure is becoming important. System monitoring data is multidimensional time series data. When we deal with multidimensional time series data, we have difficulty in considering both characteristics of multidimensional data and characteristics of time series data. When dealing with multidimensional data, correlation between variables should be considered. Existing methods such as probability and linear base, distance base, etc. are degraded due to limitations called the curse of dimensions. In addition, time series data is preprocessed by applying sliding window technique and time series decomposition for self-correlation analysis. These techniques are the cause of increasing the dimension of data, so it is necessary to supplement them. The anomaly detection field is an old research field, and statistical methods and regression analysis were used in the early days. Currently, there are active studies to apply machine learning and artificial neural network technology to this field. Statistically based methods are difficult to apply when data is non-homogeneous, and do not detect local outliers well. The regression analysis method compares the predictive value and the actual value after learning the regression formula based on the parametric statistics and it detects abnormality. Anomaly detection using regression analysis has the disadvantage that the performance is lowered when the model is not solid and the noise or outliers of the data are included. There is a restriction that learning data with noise or outliers should be used. The autoencoder using artificial neural networks is learned to output as similar as possible to input data. It has many advantages compared to existing probability and linear model, cluster analysis, and map learning. It can be applied to data that does not satisfy probability distribution or linear assumption. In addition, it is possible to learn non-mapping without label data for teaching. However, there is a limitation of local outlier identification of multidimensional data in anomaly detection, and there is a problem that the dimension of data is greatly increased due to the characteristics of time series data. In this study, we propose a CMAE (Conditional Multimodal Autoencoder) that enhances the performance of anomaly detection by considering local outliers and time series characteristics. First, we applied Multimodal Autoencoder (MAE) to improve the limitations of local outlier identification of multidimensional data. Multimodals are commonly used to learn different types of inputs, such as voice and image. The different modal shares the bottleneck effect of Autoencoder and it learns correlation. In addition, CAE (Conditional Autoencoder) was used to learn the characteristics of time series data effectively without increasing the dimension of data. In general, conditional input mainly uses category variables, but in this study, time was used as a condition to learn periodicity. The CMAE model proposed in this paper was verified by comparing with the Unimodal Autoencoder (UAE) and Multi-modal Autoencoder (MAE). The restoration performance of Autoencoder for 41 variables was confirmed in the proposed model and the comparison model. The restoration performance is different by variables, and the restoration is normally well operated because the loss value is small for Memory, Disk, and Network modals in all three Autoencoder models. The process modal did not show a significant difference in all three models, and the CPU modal showed excellent performance in CMAE. ROC curve was prepared for the evaluation of anomaly detection performance in the proposed model and the comparison model, and AUC, accuracy, precision, recall, and F1-score were compared. In all indicators, the performance was shown in the order of CMAE, MAE, and AE. Especially, the reproduction rate was 0.9828 for CMAE, which can be confirmed to detect almost most of the abnormalities. The accuracy of the model was also improved and 87.12%, and the F1-score was 0.8883, which is considered to be suitable for anomaly detection. In practical aspect, the proposed model has an additional advantage in addition to performance improvement. The use of techniques such as time series decomposition and sliding windows has the disadvantage of managing unnecessary procedures; and their dimensional increase can cause a decrease in the computational speed in inference.The proposed model has characteristics that are easy to apply to practical tasks such as inference speed and model management.

• Progress in Medical Physics
• /
• v.18 no.3
• /
• pp.179-185
• /
• 2007

The purpose of this study is to investigate the spectra of a magnetic resonance spectroscopy (MRS) in accordance with the variance of TE and the volumes of metabolites in a localized voxel for the quality assurance using a designed single voxel spectroscopy QA phantom. Because a cone-shade phantom is designed as the volume of metabolite in a localized voxel is changeable, we try to analyze the peaks of each metabolite (NAA, Cr, Cho, Lac, etc.) in accordance with metabolite volume in a localized voxel as well as echo time (TE). All data were obtained using a 3T MRI/MRS machine and analyzed using $jMRUI^{(R)}$. The results of this study show that TE is in inverse proportion to the noise of MRS and the longer TE and the less metabolite volume in the localized voxel, the peak intensities of each metabolite decrease. In case of the lactate, its peak was observed on the all TE only if the greatest metabolite is included in the localized voxel. Then, the intensity of a metabolite is more sensitive to the metabolite volume in the localized voxel than the TE. These obtained in vitro MRS data is provide the guideline that is important for in vivo metabolite quantification. But, in the edge of cone-shape vial air bubbles were observed and spectrum could not obtained. Therefore our cone-shape MRS phantom needs to be modified in order to solve these problems.

• Progress in Medical Physics
• /
• v.20 no.4
• /
• pp.244-252
• /
• 2009

The purpose of this study is to compensate for susceptibility and a ferromagnetic body artifact using CFA and TGV on MR Imaging. A total of 30 patients (15 men and 15 women, mean age: 45 years) were performed on head and neck diseases. MR Unit used a 1.5T superconducting magnet (GE medical system, High Density). This study have investigated by changing with CFA and TGV (70, 90, 110, 130, 150) searching for compensation values about susceptibility and a ferromagnetic body artifact in 60 kg standards of body weight (p<0.05). As a quality results, Image qualities were obtained at different score from CFA and TGV (70, 90, 110, 130, $150=3.23{\pm}0.35$, $4.31{\pm}0.02$ $4.23{\pm}0.21$, $5.12{\pm}0.25$, $7.13{\pm}0.72$, $8.31{\pm}0.01$, $5.21{\pm}0.15$, $6.14{\pm}0.08$, $5.23{\pm}0.72$, $5.91{\pm}0.06$, p<0.05). Absolute CNRs (TG, CNRpre, CNRpost) were acquired with (70:$-1.44{\pm}0.11$, $-2.7{\pm}0.04$, 90:$-2.18{\pm}0.42$, $-4.41{\pm}0.43$, 110:$-2.89{\pm}0.43$, $-5.23{\pm}0.02$, 130:$-2.34{\pm}0.05$, $-5.26{\pm}0.01$, 150: $-2.09{\pm}0.08$, $-3.87{\pm}0.12$, p<0.05). In conclusions, this study could be compensated for metal and flow artifacts surrounding the tissues having artifact by changing CFA and TGV.

• Korean Journal of Cognitive Science
• /
• v.3 no.1
• /
• pp.61-78
• /
• 1991

This paper descibes the study of application of a modified Neocognitron model with backward path for the recognition of Hangul(Korean) syllabic characters. In this original report, Fukushima demonstrated that Neocognitron can recognize hand written numerical characters of $19{\times}19$ size. This version accepts $61{\times}61$ images of handwritten Hangul syllabic characters or a part thereof with a mouse or with a scanner. It consists of an input layer and 3 pairs of Uc layers. The last Uc layer of this version, recognition layer, consists of 24 planes of $5{\times}5$ cells which tell us the identity of a grapheme receiving attention at one time and its relative position in the input layer respectively. It has been trained 10 simple vowel graphemes and 14 simple consonant graphemes and their spatial features. Some patterns which are not easily trained have been trained more extrensively. The trained nerwork which can classify indivisual graphemes with possible deformation, noise, size variance, transformation or retation wre then used to recongnize Korean syllabic characters using its selective attention mechanism for image segmentation task within a syllabic characters. On initial sample tests on input characters our model could recognize correctly up to 79%of the various test patterns of handwritten Korean syllabic charactes. The results of this study indeed show Neocognitron as a powerful model to reconginze deformed handwritten charavters with big size characters set via segmenting its input images as recognizable parts. The same approach may be applied to the recogition of chinese characters, which are much complex both in its structures and its graphemes. But processing time appears to be the bottleneck before it can be implemented. Special hardware such as neural chip appear to be an essestial prerquisite for the practical use of the model. Further work is required before enabling the model to recognize Korean syllabic characters consisting of complex vowels and complex consonants. Correct recognition of the neighboring area between two simple graphemes would become more critical for this task.

• Journal of radiological science and technology
• /
• v.26 no.2
• /
• pp.57-61
• /
• 2003

The purpose of this study is to compare both 1.5T and 4.7T in Praietal White matter material Phantom using the same methodology at both field strengths. Data at both field strengths are compared in terms of $T_2$ relaxation times, line widths and SNRs MR imaging and $^1H$ MR spectroscopy were performed on GE 1.5T SIGNA system and Broker Biospec 4.7T/30 MRI/MRS system. After phantom axial scan $^1H$ MRS was obtained from T2 weighted image by 3-dimensional localization technique(PRESS : Point RE solved spectroscopy Sequence) this phantom is composed of an aqueous solution 36.7 mmol/L of NAA, 25.0 mmol/L of Cr, 6.3 mmol/L of choline chloride, 30.0 mmol/L or Glu, and 22.5 mmol/L of MI(adjusted to a pH of 7,15 in a phosphate buffet). Data processed using software developed inhouse. At 1.5T, T2 relaxation times for Cho, Cr, and NAA were $0.41{\pm}0.07,\;0.26{\pm}0.04,\;0.46{\pm}0.07$ while at 4.7T they were $0.17{\pm}0.03,\;0.14{\pm}0.05,\;0.20{\pm}0.03$ respectively. At 1.5T, line widths for water, Cho, Cr and NAA were $2.9{\pm}0.7,\;1.6{\pm}0.7,\;1.7{\pm}0.8,\;2.2{\pm}0.02Hz$ while at 4.7T they were $5.2{\pm}1.1,\;4.6{\pm}1.9,\;4.01{\pm}1.8,\;4.8{\pm}1.9Hz$ respectively. It can be seen that $T_2$ relaxation times were significantly shorter at 4.7 compared to 1.5T and that the line widths were also broader. The average SNRs for NAA for subjects at short and long TEs were $23.5{\pm}11.3$ at TE=20 msec ; $15.4{\pm}7.7$ at TE=272 msec at 1.5T and $40{\pm}8.3$ and $17{\pm}3.5$ respectively at 4.7T higher field strength is superior because of improved sensitivity and chemical shift dispersion. However these improvements are partially offset by increased line widths and decrease $T_2$ relaxation times, which act to reduce both sensitivity and resolution. In our experiments with the equipment available to us, 4.7T proton spectra at short TEs exhibit moderately improved sensitivity compared to 1.5T.

• Investigative Magnetic Resonance Imaging
• /
• v.5 no.2
• /
• pp.138-148
• /
• 2001

Purpose : Within a clinically acceptable time frame, we obtained the high resolution MR images of the human brain, knee, foot and wrist from 3T whole-body MRI system which was equipped with the world first 37 active shield magnet. Materials and Methods : Spin echo (SE) and Fast Spin Echo (FSE) images were obtained from the human brain, knee, foot and wrist of normal subjects using a homemade birdcage and transverse electromagnetic (TEM) resonators operating in quadrature and tuned to 128 MHz. For acquisition of MR images of knee, foot and wrist, we employed a homemade saddle shaped RF coil. Topical common acquisition parameters were as follows: matrix=$512{\times}512$, field of view (FOV) =20 cm, slice thickness = 3 mm, number of excitations (NEX)=1. For T1-weighted MR images, we used TR = 500 ms, TE = 10 or 17.4 ms. For T2-weighted MR images, we used TR=4000 ms, TE = 108 ms. Results : Signal to noise ratio (SNR) of 3T system was measured 2.7 times greater than that of prevalent 1.5T system. MR images obtained from 3T system revealed numerous small venous structures throughout the image plane and provided reasonable delineation between gray and white matter. Conclusion The present results demonstrate that the MR images from 3T system could provide better diagnostic quali\ulcorner of resolution and sensitivity than those of 1.5T system. The elevated SNR observed in the 3T high field magnetic resonance imaging can be utilized to acquire images with a level of resolution approaching the microscopic structural level under in vivo conditions. These images represent a significant advance in our ability to examine small anatomical features with noninvasive imaging methods.

• Progress in Medical Physics
• /
• v.21 no.2
• /
• pp.137-144
• /
• 2010

The purpose of this study was to evaluate the usefulness of reducing of craniofacial radiation dose using automatic exposure control (AEC) technique in the 64 multi-detector computed tomography (MDCT). We used SOMATOM Definition 64 multi-detector CT, and head of whole body phantom (KUPBU-50, Kyoto Kagaku CO. Ltd). The protocol were helical scan method with 120 kVp, 1 sec of rotation time, 5 mm of slice thickness and increment, 250 mm of FOV, $512{\times}512$ of matrix size, $64{\times}0.625\;mm$ of collimation, and 1 of pitch. The evaluation of dose reducing effect was compared the fixed tube current of 350 with AEC technique. The image quality was measured the noise using standard deviation of CT number. The range of craniofacial bone was to mentum end from calvaria apex, which devided three regions: calvaria~superciliary ridge (1 segment), superciliary ridge~acanthion (2 segment), and acanthion~mentum (3 segment). In the fixed tube current technique, CTDIvol was 57.7 mGy, DLP was $640.2\;mGy{\cdot}cm$ in the all regions. The AEC technique was showed that 1 segment were 30.7 mGy of CTDIvol, 340.7 $mGy{\cdot}cm$ of DLP, 2 segment were 46.5 mGy of CTDIvol, $515.0\;mGy{\cdot}cm$ of DLP, and 3 segment were 30.3 mGy of CTDIvol, $337.0\;mGy{\cdot}cm$ of DLP. The standard deviation of CT number was 2.622 with the fixed tube current technique and 3.023 with the AEC technique in the 1 segment, was 3.118 with the fixed tube current technique and 3.379 with the AEC technique in the 2 segment, was 2.670 with the fixed tube current technique and 3.186 with the AEC technique in the 3 segment. The craniofacial radiation dose using AEC Technique in the 64 MDCT was evaluated the usefulness of reducing for the eye, the parotid and thyroid with high radiation sensitivity particularly.

• Investigative Magnetic Resonance Imaging
• /
• v.14 no.1
• /
• pp.31-40
• /
• 2010

Purpose : To compare the relative values of various fast breath-hold imaging sequences for superparamagnetic iron-oxide (SPIO)-enhanced hepatic MRI for the assessment of solid focal lesions with a 3T MRI unit. Materials and Methods : 102 consecutive patients with one or more solid malignant hepatic lesions were evaluated by spoiled gradient echo (GRE) sequences with three different echo times (2.4 msec [GRE_2.4], 5.8 msec [GRE_5.8], and 10 msec [GRE_10]) for $T2^*$-weighted imaging in addition to T2-weighted turbo spin echo (TSE) sequence following intravenous SPIO injection. Image qualities of the hepatic contour, vascular landmarks and artifacts were rated by two independent readers using a four-point scale. For quantitative analysis, contrast-to-noise ratio (CNR) was measured in 170 solid focal lesions larger than 1 cm (107 hepatocellular carcinomas, nine cholangiocarcinomas and 54 metastases). Results : GRE_5.8 showed the highest mean points for hepatic contour, vascular anatomy and imaging artifact presence among all of the subjected sequences (p<0.001) and was comparable (p=0.414) with GRE_10 with regard to lesion conspicuity. The mean CNRs were significantly higher (p<0.001) in the following order: GRE_10 ($24.4{\pm}14.5$), GRE_5.8 ($14.8{\pm}9.4$), TSE ($9.7{\pm}6.3$), and GRE_2.4 ($7.9{\pm}6.4$). The mean CNRs of CCCs and metastases were higher than those of HCCs for all imaging sequences (p<0.05). Conclusion : Regarding overall performances, GRE using a moderate echo time of 5.8 msec can provide the most reliable data among the various fast breath-hold SPIO-enhanced hepatic MRI sequences at 3T unit despite the lower CNR of GRE_5.8 compared to that of GRE_10.

• The Korean Journal of Nuclear Medicine
• /
• v.29 no.4
• /
• pp.533-540
• /
• 1995

Attenuation correction is important in producing quantitative positron emission tomography (PET) images. Conventionally, photon attenuation effects are corrected using transmission measurements performed before tracer administration. The pre-injection transmission measurement approach may require a time delay between transmission and emission scans for the tracer studies requiring a long uptake period, about 45 minutes for F-18 deoxyglucose study. The time delay will limit patient throughput and increase the likelihood of patient motion. A technique lot performing simultaneous transmission and emission scans (T+E method) after the tracer injection has been validated. The T+E method substracts the emission counts contaminating the transmission measurements to produce accurate attenuation correction coefficients. This method has been evaluated in experiments using a cylindrical phantom filled with background water (5750 cc) containing $0.4{\mu}Ci/cc$ of F-18 fluoride ion and one insert cylinder (276 cc) containing $4.3{\mu}Ci/cc$. GE $Advance^{TM}$ PET scanner and Ge-68 rotating pin sources for transmission scanning were used for this investigation. Post-injection transmission scan and emission scan were peformed alternatively over time. The error in emission images corrected using post-infection transmission scan to emission images corrected transmission scan was 2.6% at the concentration of $1.0{\mu}Ci/cc$. No obvious differences in image quality and noise were apparent between the two images. The attenuation correction can be accomplished with post-injection transmission measurement using rotating pin sources and this method can significantly shorten the time between transmission and omission scans and thereby reduce the likelihood of patient motion and increase scanning throughput in PET.

• The Korean Journal of Nuclear Medicine Technology
• /
• v.18 no.1
• /
• pp.127-133
• /
• 2014

Purpose: Low dose of PET/CT is important because of Patient's X-ray exposure. The aim of this study was to evaluate the effectiveness of low-dose PET/ CT image through the CTAC and QAC of patient study and phantom study. Materials and Methods: We used the discovery 710 PET/CT (GE). We used the NEMA IEC body phantom for evaluating the PET data corrected by ultra-low dose CT attenuation correction method and NU2-94 phantom for uniformity. After injection of 70.78 MBq and 22.2 MBq of 18 F-FDG were done to each of phantom, PET/CT scans were obtained. PET data were reconstructed by using of CTAC of which dose was for the diagnosis CT and Q. AC of which was only for attenuation correction. Quantitative analysis was performed by use of horizontal profile and vertical profile. Reference data which were corrected by CTAC were compared to PET data which was corrected by the ultra-low dose. The relative error was assessed. Patients with over weighted and normal weight also underwent a PET/CT scans according to low dose protocol and standard dose protocol. Relative error and signal to noise ratio of SUV were analyzed. Results: In the results of phantom test, phantom PET data were corrected by CTAC and Q.AC and they were compared each other. The relative error of Q.AC profile was been calculated, and it was shown in graph. In patient studies, PET data for overweight patient and normal weight patient were reconstructed by CTAC and Q.AC under routine dose and ultra-low dose. When routine dose was used, the relative error was small. When high dose was used, the result of overweight patient was effectively corrected by Q.AC. Conclusion: In phantom study, CTAC method with 80 kVp and 10 mA was resulted in bead hardening artifact. PET data corrected by ultra- low dose CTAC was not quantified, but those by the same dose were quantified properly. In patients' cases, PET data of over weighted patient could be quantified by Q.AC method. Its relative difference was not significant. Q.AC method was proper attenuation correction method when ultra-low dose was used. As a result, it is expected that Q.AC is a good method in order to reduce patient's exposure dose.