• Journal of the Korean Geophysical Society
• /
• v.2 no.1
• /
• pp.39-44
• /
• 1999

High-resolution images are drawn from existing seismic data which were originally obtained by Korea Ocean Research & Development Institute (KORDI) during 1994-1997 for deep seismic studies on the East Sea of Korea. These images are analyzed for mapping Quaternary faults and near-bottom gas pockets. First 12 channels are selected from shot gathers for reprocessing. The processing sequence adopted for high-resolution seismic images comprises data copy, trace editing, true amplitude recovery, common-midpoint sorting, initial muting, prestack deconvolution, bandpass filtering, stacking, highpass filtering, poststack deconvolution, f-x migration, and automatic gain control (AGC). Among these processing steps, predictive deconvolution, highpass filtering, and short window AGC are the most significant in enhancement of resolution. More than 200 Quaternanry faults are interpreted on the migrated sections in the shallow depths beneath the seafloor. Although numerous faults are found mostly at the western continental slope and boundaries of the Ulleung Basin, significant amount of the faults are also indicated within the basin. Many of these faults are believed to be formed with reactivation of basement, from geotectonic activities including volcanism, and often originated in Tertiary, indicating that the tectonic regime of the East Sea might be unstable. Existence of shallow gas pockets casts real hazardous warnings to deep-sea drillings and/or to underwater constructions such as inter-island cables and gas pipelines. On the other hand, discovery of these gas pockets heightens the interests in developing natural resources in the East Sea. Reprocessed seismic sections, however, show no typical seismic characteristics for gas hydrates such as bottom-simulating reflectors in the western continental slope and ocean floor.

• Journal of the Korean Society of Radiology
• /
• v.83 no.3
• /
• pp.669-679
• /
• 2022

Purpose To evaluate the feasibility of pediatric low-dose facial CT reconstructed with filtered back projection (FBP) using adequate kernels. Materials and Methods We retrospectively reviewed the clinical and imaging data of children aged < 10 years who underwent facial CT at our emergency department. The patients were divided into two groups: low-dose CT (LDCT; Group A, n = 73) with a fixed 80-kVp tube potential and automatic tube current modulation (ATCM) and standard-dose CT (SDCT; Group B, n = 40) with a fixed 120-kVp tube potential and ATCM. All images were reconstructed with FBP using bone and soft tissue kernels in Group A and only bone kernel in Group B. The groups were compared in terms of image noise, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR). Two radiologists subjectively scored the overall image quality of bony and soft tissue structures. The CT dose index volume and dose-length product were recorded. Results Image noise was higher in Group A than in Group B in bone kernel images (p < 0.001). Group A using a soft tissue kernel showed the highest SNR and CNR for all soft tissue structures (all p < 0.001). In the qualitative analysis of bony structures, Group A scores were found to be similar to or higher than Group B scores on comparing bone kernel images. In the qualitative analysis of soft tissue structures, there was no significant difference between Group A using a soft tissue kernel and Group B using a bone kernel with a soft tissue window setting (p > 0.05). Group A showed a 76.9% reduction in radiation dose compared to Group B (3.2 ± 0.2 mGy vs. 13.9 ± 1.5 mGy; p < 0.001). Conclusion The addition of a soft tissue kernel image to conventional CT reconstructed with FBP enables the use of pediatric low-dose facial CT protocol while maintaining image quality.

• Progress in Medical Physics
• /
• v.20 no.4
• /
• pp.298-307
• /
• 2009

By using the MR T2 map technique, this study intends, first, to measure the change of T2 values of cartilage between healthy people and patients with osteoarthritis and, second, to assess the form and the damage of cartilage in the knee-joint, through which this study would consider the utility of the T2 map technique. Thirty healthy people were selected based on their clinical history and current status and another thirty patients with osteoarthritis of the knee who were screened by simple X-ray from November 2007 to December 2008 were selected. Their T2 Spin Echo (SE hereafter) images for the cartilage of the knee joint were collected by using the T2 SE sequence, one of the multi-echo methods (TR: 1,000 ms; TE values: 6.5, 13, 19.5, 26, 32.5. 40, 45.5, 52). Based on these images, the changes in the signal intensity (SI hereafter) for each section of the cartilage of the knee joint were measured, which yielded average values of T2 through the Origin 7.0 Professional (Northampton, MA 01060 USA). With these T2s, the independent samples T-test was performed by SPSS Window version 12.0 to run the quantitative analysis and to test the statistical significance between the healthy group and the patient group. Closely looking at T2 values for each anterior and lateral articular cartilage of the sagittal plane and the coronal plane, in the sagittal plane, the average T2 of the femoral cartilage in the patient group with arthritis of the knee ($42.22{\pm}2.91$) was higher than the average T2 of the healthy group ($36.26{\pm}5.01$). Also, the average T2 of the tibial cartilage in the patient group ($43.83{\pm}1.43$) was higher than the average T2 in the healthy group ($36.45{\pm}3.15$). In the case of the coronal plane, the average T2 of the medial femoral cartilage in the patient group ($45.65{\pm}7.10$) was higher than the healthy group ($36.49{\pm}8.41$) and so did the average T2 of the anterior tibial cartilage (i.e., $44.46{\pm}3.44$ for the patient group vs. $37.61{\pm}1.97$ for the healthy group). As for the lateral femoral cartilage in the coronal plane, the patient group displayed the higher T2 ($43.41{\pm}4.99$) than the healthy group did ($37.64{\pm}4.02$) and this tendency was similar in the lateral tibial cartilage (i.e., $43.78{\pm}8.08$ for the patient group vs. $36.62{\pm}7.81$ for the healthy group). Along with the morphological MR imaging technique previously used, the T2 map technique seems to help patients with cartilage problems, in particular, those with the arthritis of the knee for early diagnosis by quantitatively analyzing the structural and functional changes of the cartilage.

• Journal of Intelligence and Information Systems
• /
• v.25 no.1
• /
• pp.163-177
• /
• 2019

As smartphones are getting widely used, human activity recognition (HAR) tasks for recognizing personal activities of smartphone users with multimodal data have been actively studied recently. The research area is expanding from the recognition of the simple body movement of an individual user to the recognition of low-level behavior and high-level behavior. However, HAR tasks for recognizing interaction behavior with other people, such as whether the user is accompanying or communicating with someone else, have gotten less attention so far. And previous research for recognizing interaction behavior has usually depended on audio, Bluetooth, and Wi-Fi sensors, which are vulnerable to privacy issues and require much time to collect enough data. Whereas physical sensors including accelerometer, magnetic field and gyroscope sensors are less vulnerable to privacy issues and can collect a large amount of data within a short time. In this paper, a method for detecting accompanying status based on deep learning model by only using multimodal physical sensor data, such as an accelerometer, magnetic field and gyroscope, was proposed. The accompanying status was defined as a redefinition of a part of the user interaction behavior, including whether the user is accompanying with an acquaintance at a close distance and the user is actively communicating with the acquaintance. A framework based on convolutional neural networks (CNN) and long short-term memory (LSTM) recurrent networks for classifying accompanying and conversation was proposed. First, a data preprocessing method which consists of time synchronization of multimodal data from different physical sensors, data normalization and sequence data generation was introduced. We applied the nearest interpolation to synchronize the time of collected data from different sensors. Normalization was performed for each x, y, z axis value of the sensor data, and the sequence data was generated according to the sliding window method. Then, the sequence data became the input for CNN, where feature maps representing local dependencies of the original sequence are extracted. The CNN consisted of 3 convolutional layers and did not have a pooling layer to maintain the temporal information of the sequence data. Next, LSTM recurrent networks received the feature maps, learned long-term dependencies from them and extracted features. The LSTM recurrent networks consisted of two layers, each with 128 cells. Finally, the extracted features were used for classification by softmax classifier. The loss function of the model was cross entropy function and the weights of the model were randomly initialized on a normal distribution with an average of 0 and a standard deviation of 0.1. The model was trained using adaptive moment estimation (ADAM) optimization algorithm and the mini batch size was set to 128. We applied dropout to input values of the LSTM recurrent networks to prevent overfitting. The initial learning rate was set to 0.001, and it decreased exponentially by 0.99 at the end of each epoch training. An Android smartphone application was developed and released to collect data. We collected smartphone data for a total of 18 subjects. Using the data, the model classified accompanying and conversation by 98.74% and 98.83% accuracy each. Both the F1 score and accuracy of the model were higher than the F1 score and accuracy of the majority vote classifier, support vector machine, and deep recurrent neural network. In the future research, we will focus on more rigorous multimodal sensor data synchronization methods that minimize the time stamp differences. In addition, we will further study transfer learning method that enables transfer of trained models tailored to the training data to the evaluation data that follows a different distribution. It is expected that a model capable of exhibiting robust recognition performance against changes in data that is not considered in the model learning stage will be obtained.