• Journal of Intelligence and Information Systems
• /
• v.22 no.2
• /
• pp.127-142
• /
• 2016

Deep learning model is a kind of neural networks that allows multiple hidden layers. There are various deep learning architectures such as convolutional neural networks, deep belief networks and recurrent neural networks. Those have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. Among those architectures, convolutional neural networks and recurrent neural networks are classified as the supervised learning model. And in recent years, those supervised learning models have gained more popularity than unsupervised learning models such as deep belief networks, because supervised learning models have shown fashionable applications in such fields mentioned above. Deep learning models can be trained with backpropagation algorithm. Backpropagation is an abbreviation for "backward propagation of errors" and a common method of training artificial neural networks used in conjunction with an optimization method such as gradient descent. The method calculates the gradient of an error function with respect to all the weights in the network. The gradient is fed to the optimization method which in turn uses it to update the weights, in an attempt to minimize the error function. Convolutional neural networks use a special architecture which is particularly well-adapted to classify images. Using this architecture makes convolutional networks fast to train. This, in turn, helps us train deep, muti-layer networks, which are very good at classifying images. These days, deep convolutional networks are used in most neural networks for image recognition. Convolutional neural networks use three basic ideas: local receptive fields, shared weights, and pooling. By local receptive fields, we mean that each neuron in the first(or any) hidden layer will be connected to a small region of the input(or previous layer's) neurons. Shared weights mean that we're going to use the same weights and bias for each of the local receptive field. This means that all the neurons in the hidden layer detect exactly the same feature, just at different locations in the input image. In addition to the convolutional layers just described, convolutional neural networks also contain pooling layers. Pooling layers are usually used immediately after convolutional layers. What the pooling layers do is to simplify the information in the output from the convolutional layer. Recent convolutional network architectures have 10 to 20 hidden layers and billions of connections between units. Training deep learning networks has taken weeks several years ago, but thanks to progress in GPU and algorithm enhancement, training time has reduced to several hours. Neural networks with time-varying behavior are known as recurrent neural networks or RNNs. A recurrent neural network is a class of artificial neural network where connections between units form a directed cycle. This creates an internal state of the network which allows it to exhibit dynamic temporal behavior. Unlike feedforward neural networks, RNNs can use their internal memory to process arbitrary sequences of inputs. Early RNN models turned out to be very difficult to train, harder even than deep feedforward networks. The reason is the unstable gradient problem such as vanishing gradient and exploding gradient. The gradient can get smaller and smaller as it is propagated back through layers. This makes learning in early layers extremely slow. The problem actually gets worse in RNNs, since gradients aren't just propagated backward through layers, they're propagated backward through time. If the network runs for a long time, that can make the gradient extremely unstable and hard to learn from. It has been possible to incorporate an idea known as long short-term memory units (LSTMs) into RNNs. LSTMs make it much easier to get good results when training RNNs, and many recent papers make use of LSTMs or related ideas.

• Journal of the Korean Society of Radiology
• /
• v.13 no.5
• /
• pp.801-809
• /
• 2019

Existing Gamma Knife Radiosurgery(GKRS) for large lesions is often conducted in stages with volume or dose partitions. Often in case of volume division the target used to be divided into sub-volumes which are irradiated under the determined prescription dose in multi-sessions separated by a day or two, 3~6 months. For the entire course of treatment, treatment informations of the previous stages needs to be reflected to subsequent sessions on the newly mounted stereotactic frame through coordinate transformation between sessions. However, it is practically difficult to implement the previous dose distributions with existing Gamma Knife system except in the same stereotactic space. The treatment area is expanding because it is possible to perform the multistage treatment using the latest Gamma Knife Platform(GKP). The purpose of this study is to introduce the image-coregistration based on the stereotactic spaces and the strategy of multistage GKRS such as the determination of prescription dose at each stage using new GKP. Usually in image-coregistration either surgically-embedded fiducials or internal anatomical landmarks are used to determine the transformation relationship. Author compared the accuracy of coordinate transformation between multi-sessions using four or six anatomical landmarks as an example using internal anatomical landmarks. Transformation matrix between two stereotactic spaces was determined using PseudoInverse or Singular Value Decomposition to minimize the discrepancy between measured and calculated coordinates. To evaluate the transformation accuracy, the difference between measured and transformed coordinates, i.e., ${\Delta}r$, was calculated using 10 landmarks. Four or six points among 10 landmarks were used to determine the coordinate transformation, and the rest were used to evaluate the approaching method. Each of the values of ${\Delta}r$ in two approaching methods ranged from 0.6 mm to 2.4 mm, from 0.17 mm to 0.57 mm. In addition, a method of determining the prescription dose to give the same effect as the treatment of the total lesion once in case of lesion splitting was suggested. The strategy of multistage treatment in the same stereotactic space is to design the treatment for the whole lesion first, and the whole treatment design shots are divided into shots of each stage treatment to construct shots of each stage and determine the appropriate prescription dose at each stage. In conclusion, author confirmed the accuracy of prescribing dose determination as a multistage treatment strategy and found that using as many internal landmarks as possible than using small landmarks to determine coordinate transformation between multi-sessions yielded better results. In the future, the proposed multistage treatment strategy will be a great contributor to the frameless fractionated treatment of several Gamma Knife Centers.

• Korean Journal of Remote Sensing
• /
• v.39 no.6_1
• /
• pp.1321-1330
• /
• 2023

Normalized Difference Vegetation Index (NDVI) is the most widely used remote sensing data in the agricultural field and is currently provided by most optical satellites. In particular, as high-resolution optical satellite images become available, the selection of optimal optical satellite images according to agricultural applications has become a very important issue. In this study, we aim to define the most optimal optical satellite image when monitoring NDVI in rice fields in Korea and derive the resolution-related requirements necessary for this. For this purpose, we compared and analyzed the spatial distribution and time series patterns of the Dangjin rice paddy in Korea from 2019 to 2022 using NDVI images from MOD13, Landsat-8, Sentinel-2A/B, and PlanetScope satellites, which are widely used around the world. Each data is provided with a spatial resolution of 3 m to 250 m and various periods, and the area of the spectral band used to calculate NDVI also has slight differences. As a result of the analysis, Landsat-8 showed the lowest NDVI value and had very low spatial variation. In comparison, the MOD13 NDVI image showed similar spatial distribution and time series patterns as the PlanetScope data but was affected by the area surrounding the rice field due to low spatial resolution. Sentinel-2A/B showed relatively low NDVI values due to the wide near-infrared band area, and this feature was especially noticeable in the early stages of growth. PlanetScope's NDVI provides detailed spatial variation and stable time series patterns, but considering its high purchase price, it is considered to be more useful in small field areas than in spatially uniform rice paddy. Accordingly, for rice field areas, 250 m MOD13 NDVI or 10 m Sentinel-2A/B are considered to be the most efficient, but high-resolution satellite images can be used to estimate detailed physical quantities of individual crops.

• The Sea:JOURNAL OF THE KOREAN SOCIETY OF OCEANOGRAPHY
• /
• v.15 no.1
• /
• pp.41-50
• /
• 2010

We have developed an in-situ benthic chamber (BelcI) for use in coastal studies that can be deployed from a small boat. It is expected that BelcI will be useful in studying the benthic boundary layer because of its flexibility. BelcI is divided into three main areas: 1) frame and body chamber, 2) water sampler, and 3) stirring devices, electric controller, and data acquisition technology. To maximize in-situ use, the frame is constructed from two layers that consist of square cells. All electronic parts (motor controller, pA meter, data acquisition, etc.) are low-power consumers so that the external power supply can be safely removed from the system. The hydrodynamics of BelcI, measured by PIV (particle image velocimetry), show a typical "radial-flow impeller" pattern. Mixing time of water in the chamber is about 30 s, and shear velocity ($u^*$) near the bottom layer was calculated at $0.32\;cm\;s^{-1}$. Measurements of diffusivity boundary layer thickness showed a range of $180-230\;{\mu}m$. Sediment oxygen consumption rate, measured in-situ，was $84\;mmol\;O_2\;m^{-2}\;d_{-1}$, more than two times higher than on-board incubation results. Benthic fluxes assessed from in-situ incubation were estimated as follows: nitrate + nitrite = $0.18\;{\pm}\;0.07\;mmol\;m^{-2}\;d^{-1}$ ammonium $23\;{\pm}\;1\;mmol\;m^{-2}\;d^{-1}$ phosphate = $0.09\;{\pm}\;0.02\;mmol\;m^{-2}\;d^{-1}$ and silicate = $23\;{\pm}\;1\;mmol\;m^{-2}\;d^{-1}$.

• Journal of the Korean Society of Radiology
• /
• v.18 no.1
• /
• pp.45-52
• /
• 2024

In this study, an AI-based algorithm was developed to prevent image quality deterioration and reading errors due to patient movement in PET/CT examinations that use radioisotopes in medical institutions to test cancer and other diseases. Using the Mothion Free software developed using, we checked the degree of correction of movement due to breathing, evaluated its usefulness, and conducted a study for clinical application. The experimental method was to use an RPM Phantom to inject the radioisotope 18F-FDG into a vacuum vial and a sphere of a NEMA IEC body Phantom of different sizes, and to produce images by directing the movement of the radioisotope into a moving lesion during respiration. The vacuum vial had different degrees of movement at different positions, and the spheres of the NEMA IEC body Phantom of different sizes produced different sizes of lesions. Through the acquired images, the lesion volume, maximum SUV, and average SUV were each measured to quantitatively evaluate the degree of motion correction by Motion Free. The average SUV of vacuum vial A, with a large degree of movement, was reduced by 23.36 %, and the error rate of vacuum vial B, with a small degree of movement, was reduced by 29.3 %. The average SUV error rate at the sphere 37mm and 22mm of the NEMA IEC body Phantom was reduced by 29.3 % and 26.51 %, respectively. The average error rate of the four measurements from which the error rate was calculated decreased by 30.03 %, indicating a more accurate average SUV value. In this study, only two-dimensional movements could be produced, so in order to obtain more accurate data, a Phantom that can embody the actual breathing movement of the human body was used, and if the diversity of the range of movement was configured, a more accurate evaluation of usability could be made.

• Journal of Internet Computing and Services
• /
• v.25 no.1
• /
• pp.99-107
• /
• 2024

Along with economic growth and industrial development, there is an increasing demand for various electronic components and device production of semiconductor, SMT component, and electrical battery products. However, these products may contain foreign substances coming from manufacturing process such as iron, aluminum, plastic and so on, which could lead to serious problems or malfunctioning of the product, and fire on the electric vehicle. To solve these problems, it is necessary to determine whether there are foreign materials inside the product, and may tests have been done by means of non-destructive testing methodology such as ultrasound ot X-ray. Nevertheless, there are technical challenges and limitation in acquiring X-ray images and determining the presence of foreign materials. In particular Small-sized or low-density foreign materials may not be visible even when X-ray equipment is used, and noise can also make it difficult to detect foreign objects. Moreover, in order to meet the manufacturing speed requirement, the x-ray acquisition time should be reduced, which can result in the very low signal- to-noise ratio(SNR) lowering the foreign material detection accuracy. Therefore, in this paper, we propose a five-step approach to overcome the limitations of low resolution, which make it challenging to detect foreign substances. Firstly, global contrast of X-ray images are increased through histogram stretching methodology. Second, to strengthen the high frequency signal and local contrast, we applied local contrast enhancement technique. Third, to improve the edge clearness, Unsharp masking is applied to enhance edges, making objects more visible. Forth, the super-resolution method of the Residual Dense Block (RDB) is used for noise reduction and image enhancement. Last, the Yolov5 algorithm is employed to train and detect foreign objects after learning. Using the proposed method in this study, experimental results show an improvement of more than 10% in performance metrics such as precision compared to low-density images.

• The Journal of Korean Society for Radiation Therapy
• /
• v.29 no.2
• /
• pp.101-108
• /
• 2017

Purpose: The proton used in proton therapy has a characteristic of giving a small dose to the normal tissue in front of the tumor site while forming a Bragg peak at the cancer tissue site and giving up the maximum dose and disappearing immediately. It is very important to verify the proton arrival position. In this study, we used the off-line PET CT method to measure the distribution of positron emitted from nucleons such as 11C (half-life = 20 min), 150 (half-life = 2 min) and 13N The range and distal falloff point of the proton were verified by measurement. Materials and Methods: In the IEC 2001 Body Phantom, 37 mm, 28 mm, and 22 mm spheres were inserted. The phantom was filled with water to obtain a CT image for each sphere size. To verify the proton range and distal falloff points, As a treatment planning system, SOBP were set at 46 mm on 37 mm sphere, 37 mm on 28 mm, and 33 mm on 22 mm sphere for each sphere size. The proton was scanned in the same center with a single beam of Gantry 0 degree by the scanning method. The phantom was scanned using PET-CT equipment. In the PET-CT image acquisition method, 50 images were acquired per minute, four ROIs including the spheres in the phantom were set, and 10 images were reconstructed. The activity profile according to the depth was compared to the dose profile according to the sphere size established in the treatment plan Results: The PET-CT activity profile decreased rapidly at the distal falloff position in the 37 mm, 28 mm, and 22 mm spheres as well as the dose profile. However, in the SOBP section, which is a range for evaluating the range, the results in the proximal part of the activity profile are different from those of the dose profile, and the distal falloff position is compared with the proton therapy plan and PET-CT As a result, the maximum difference of 1.4 mm at the 50 % point of the Max dose, 1.1 mm at the 45 % point at the 28 mm sphere, and the difference at the 22 mm sphere at the maximum point of 1.2 mm were all less than 1.5 mm in the 37 mm sphere. Conclusion: To maximize the advantages of proton therapy, it is very important to verify the range of the proton beam. In this study, the proton range was confirmed by the SOBP and the distal falloff position of the proton beam using PET-CT. As a result, the difference of the distally falloff position between the activity distribution measured by PET-CT and the proton therapy plan was 1.4 mm, respectively. This may be used as a reference for the dose margin applied in the proton therapy plan.

• Investigative Magnetic Resonance Imaging
• /
• v.5 no.1
• /
• pp.43-48
• /
• 2001

Purpose : To evaluate changes in rabbit liver parenchyma on MR images following percutaneous Holmium-166 injection, and to correlate those changes with histologic findings. Materials and methods. Holmium-166 (10-25 mCi) was percutaneously injected into the liver of rabbit (n=12) under sonographic guidance. MR images were obtained between one to two weeks (acute phasea) after the injection in four rabbits, and between two to four weeks (subacute phase) after the injection in four rabbits. Tissue specimens of these eight rabbits were obtained immediately after MR imaging. Tissue specimens were obtained without MR imaging in four rabbits (between one to two weeks in one rabbit and between three to four weeks in three rabbits). Results : Tissue specimens showed central liquefactive necrosis and peripheral coagulative necrosis containing deposition of small particles and hemorrhage. The peripheral margin of the lesions showed formation of the granulation tissue with fibrosis, which tended to be more prominent in subacute phase. The area of the necrosis tended to correlate with the dose of the radioactive Holmium-166. On MR images, the central portion of the necrosis showed hyperintensity on 72-weighted image, hypointensity on the precontrast T1-weighted images, and no enhancement on the dynamic MR images. The peripheral portion of the necrosis showed hypointensity on T2-weighted images, iso or mild hypointensity on the T1-weighted images, and mild peripheral enhancement on the delayed dynamic MR images. The peripheral margin of the lesion showed hypointensity on both T1- and T1-weighted images with increased enhancement on the delayed phase images of the dynamic MR images. Conclusion : After percutaneous Holmium-166 injection into rabbit liver parenchyma, the central portion showed liquefactive necrosis, the peripheral portion showed coagulative necrosis with granulation, fibrosis, hemorrhage and depostition of small granules. MR imaging may be helpful in evaluation of the histological change of the liver after percutaneous Holmium-166 treatment.

• Korean Journal of Animal Reproduction
• /
• v.21 no.2
• /
• pp.147-156
• /
• 1997

Ultrasound-guided follicular aspiration was performed in Holstein heifers once weekly with or without pretreatment of single or multiple decreasing doses using a total of 400 mg FSH. Oocytes were aspirated with a 6.5 MHz convex-array ultrasound trasducer designed for intravaginal use. All the visible follicles larger than 4 mm in diameter were punctured with a 17 gauge, 55 cm needle at each aspiration session and the follicular fluids containing oocytes were obtained by vacuum suction. The results obtained were as follows: As a preliminary experiment, the recovery rates of folicular oocytes by ultrasound-guided aspiration from the isolated ovaries of Korean native cows were compared between suction methods using manual syringe or vacuum pump. The recovery rate of oocytes using vacuum pump (80.7%) was significantly (P<0.05) higher than that using manual syringe (47.1%). The follicles were counted by their size in diameter with ultrasound image, and recovery rates and grades of follicular oocytes collected by ultrasound-guided aspiration were investigated in Holstein heifers pretreated with or without FSH. A group of heifiers were injected with multiple decreasing doses (twice a day for 3 days) of a total of 400 mg FSH. The other 2 groups were injected with a single dose of 400 mg FSH mixed with 25% PVP. Ultrasound observation of follicle population and/or ultrasound-guided transvaginal oocyte aspiration were performed 12 hrs following the last FSH injection in the multiple dose group, and 48 or 60 hrs after FSH injection in the single dose groups. Most of the visible follicles had small size of less than 3 mm in diameter in unstimulated heifers (71.0%), but medium size in all the heifers treated with FSH. (70.5 to 92.8%). The number of OPU follicles per session (4.6$\pm$1.9) were much less, compared to the vilsible follicle counts (9.7$\pm$2.2), in the nustimulated heifers due to the small dominant follicles. Among 4 goups of heifers the most visible as well as OPU follicles were observed in the heifers at 60 hrs following treatment of a single dose of 400 mg FSH (21.2$\pm$2.3 and 21.0$\pm$2.0), and the differences in both the follicle counts between the groups was found significant (P<0.05) The rates of oocyte recovery from the follicles by ultrasound-guilded aspiration were varied 46.3 to 75.0% in the heifers unstimulated and treated with a single dose of 400 mg FSH, but the group difference was not significant. The number of recovered oocytes per session a, pp.ared to be highest at aspiration at 60 hrs following single FSH (10.6$\pm$2.2) than at aspiration at 48 hrs after single FSH (7.8$\pm$2.7) or in the unstimulated heifers (3.4$\pm$3.0). The proportion of grade I and II oocytes to all oocytes collected was varied 31.8 to 64.0% between the groups. However, there was found no significant difference in both the number of oocytes recovered per session and the percentage and the percentage of grade I and II oocytes. From the above results it was concluded that the more oocytes of superior quality might be recovered economically by ultrasound-guided aspiration at 60 hrs following the pretreatment of a single dose of 400 mg FSH and by suction using a vacuum pump system of about negative pressure of 75 to 85 mmHg.

• The Korean Journal of Nuclear Medicine
• /
• v.33 no.5
• /
• pp.439-451
• /
• 1999

Purpose : It is important to differentiate malignant from benign lesions of intraocular masses in choosing therapeutic plan. Biopsy of intraocular tumor is not recommended due to the risk of visual damage. We evaluated the usefulness of F-18-FDG PET imaging in diagnosing intraocular neoplasms. Materials and Methods: F-18-FDG PET scan was performed in 13 patients (15 lesions) suspected to have malignant intraocular tumors. There were 3 benign lesions (retinal detachment, choroidal effusion and hemorrhage) and 10 patients with 12 malignant lesions (3 melanomas, 7 retinoblastomas and 2 metastatic cancers). Regional eye images ($256{\times}256$ and $128{\times}128$ matrices) were obtained with or without attenuation correction. Whole body scan was also performed in eight patients (3 benign and 6 malignant lesions). Results: All malignant lesions were visualized while all benign lesions were not visualized. The mean peak standardized uptake value (SUV) of malignant lesions was $2.64{\pm}0.57g/ml$. There was no correlations between peak SUV and tumor volume. Two large malignant lesions ($> 1000 mm^3$) showed hot uptake on whole body scan. But two medium-sized lesions ($100-1000mm^3$) looked faint and two small ($<100mm^3$) lesions were not visualized. The images reconstructed with $256{\times}256$ matrix showed lesions more clearly than those with $128{\times}128$ matrix Conclusion: F-18-FDG PET scan is highly sensitivity in detecting malignant intraocular tumor For the evaluation of small-sized intraocular lesions, whole body scan is not appropriate because of low sensitivity. A regional scan with sufficient acquisition time is recommended for that purpose. Image reconstruction in matrix size of $256{\times}256$ produced clearer images than the ones in $128{\times}128$, but it does not affect the diagnostic sensitivity.