• Korean Journal of Radiology
• /
• v.25 no.4
• /
• pp.363-373
• /
• 2024

Objective: To develop and evaluate a deep learning model for automated segmentation and detection of bone metastasis on spinal MRI. Materials and Methods: We included whole spine MRI scans of adult patients with bone metastasis: 662 MRI series from 302 patients (63.5 ± 11.5 years; male:female, 151:151) from three study centers obtained between January 2015 and August 2021 for training and internal testing (random split into 536 and 126 series, respectively) and 49 MRI series from 20 patients (65.9 ± 11.5 years; male:female, 11:9) from another center obtained between January 2018 and August 2020 for external testing. Three sagittal MRI sequences, including non-contrast T1-weighted image (T1), contrast-enhanced T1-weighted Dixon fat-only image (FO), and contrast-enhanced fat-suppressed T1-weighted image (CE), were used. Seven models trained using the 2D and 3D U-Nets were developed with different combinations (T1, FO, CE, T1 + FO, T1 + CE, FO + CE, and T1 + FO + CE). The segmentation performance was evaluated using Dice coefficient, pixel-wise recall, and pixel-wise precision. The detection performance was analyzed using per-lesion sensitivity and a free-response receiver operating characteristic curve. The performance of the model was compared with that of five radiologists using the external test set. Results: The 2D U-Net T1 + CE model exhibited superior segmentation performance in the external test compared to the other models, with a Dice coefficient of 0.699 and pixel-wise recall of 0.653. The T1 + CE model achieved per-lesion sensitivities of 0.828 (497/600) and 0.857 (150/175) for metastases in the internal and external tests, respectively. The radiologists demonstrated a mean per-lesion sensitivity of 0.746 and a mean per-lesion positive predictive value of 0.701 in the external test. Conclusion: The deep learning models proposed for automated segmentation and detection of bone metastases on spinal MRI demonstrated high diagnostic performance.

• Korean Journal of Radiology
• /
• v.23 no.7
• /
• pp.720-731
• /
• 2022

Objective: We aimed to develop and test a deep learning algorithm (DLA) for fully automated measurement of the volume and signal intensity (SI) of the liver and spleen using gadoxetic acid-enhanced hepatobiliary phase (HBP)-magnetic resonance imaging (MRI) and to evaluate the clinical utility of DLA-assisted assessment of functional liver capacity. Materials and Methods: The DLA was developed using HBP-MRI data from 1014 patients. Using an independent test dataset (110 internal and 90 external MRI data), the segmentation performance of the DLA was measured using the Dice similarity score (DSS), and the agreement between the DLA and the ground truth for the volume and SI measurements was assessed with a Bland-Altman 95% limit of agreement (LOA). In 276 separate patients (male:female, 191:85; mean age ± standard deviation, 40 ± 15 years) who underwent hepatic resection, we evaluated the correlations between various DLA-based MRI indices, including liver volume normalized by body surface area (LV_BSA), liver-to-spleen SI ratio (LSSR), MRI parameter-adjusted LSSR (aLSSR), LSSR × LV_BSA, and aLSSR × LV_BSA, and the indocyanine green retention rate at 15 minutes (ICG-R15), and determined the diagnostic performance of the DLA-based MRI indices to detect ICG-R15 ≥ 20%. Results: In the test dataset, the mean DSS was 0.977 for liver segmentation and 0.946 for spleen segmentation. The Bland-Altman 95% LOAs were 0.08% ± 3.70% for the liver volume, 0.20% ± 7.89% for the spleen volume, -0.02% ± 1.28% for the liver SI, and -0.01% ± 1.70% for the spleen SI. Among DLA-based MRI indices, aLSSR × LV_BSA showed the strongest correlation with ICG-R15 (r = -0.54, p < 0.001), with area under receiver operating characteristic curve of 0.932 (95% confidence interval, 0.895-0.959) to diagnose ICG-R15 ≥ 20%. Conclusion: Our DLA can accurately measure the volume and SI of the liver and spleen and may be useful for assessing functional liver capacity using gadoxetic acid-enhanced HBP-MRI.

• Korean Journal of Radiology
• /
• v.23 no.10
• /
• pp.949-958
• /
• 2022

Objective: To investigate the feasibility of using a deep learning-based analysis of auscultation data to predict significant stenosis of arteriovenous fistulas (AVF) in patients undergoing hemodialysis requiring percutaneous transluminal angioplasty (PTA). Materials and Methods: Forty patients (24 male and 16 female; median age, 62.5 years) with dysfunctional native AVF were prospectively recruited. Digital sounds from the AVF shunt were recorded using a wireless electronic stethoscope before (pre-PTA) and after PTA (post-PTA), and the audio files were subsequently converted to mel spectrograms, which were used to construct various deep convolutional neural network (DCNN) models (DenseNet201, EfficientNetB5, and ResNet50). The performance of these models for diagnosing ≥ 50% AVF stenosis was assessed and compared. The ground truth for the presence of ≥ 50% AVF stenosis was obtained using digital subtraction angiography. Gradient-weighted class activation mapping (Grad-CAM) was used to produce visual explanations for DCNN model decisions. Results: Eighty audio files were obtained from the 40 recruited patients and pooled for the study. Mel spectrograms of "pre-PTA" shunt sounds showed patterns corresponding to abnormal high-pitched bruits with systolic accentuation observed in patients with stenotic AVF. The ResNet50 and EfficientNetB5 models yielded an area under the receiver operating characteristic curve of 0.99 and 0.98, respectively, at optimized epochs for predicting ≥ 50% AVF stenosis. However, Grad-CAM heatmaps revealed that only ResNet50 highlighted areas relevant to AVF stenosis in the mel spectrogram. Conclusion: Mel spectrogram-based DCNN models, particularly ResNet50, successfully predicted the presence of significant AVF stenosis requiring PTA in this feasibility study and may potentially be used in AVF surveillance.

• Korean Journal of Radiology
• /
• v.22 no.7
• /
• pp.1213-1224
• /
• 2021

Objective: To develop a machine learning (ML) pipeline based on radiomics to predict Coronavirus Disease 2019 (COVID-19) severity and the future deterioration to critical illness using CT and clinical variables. Materials and Methods: Clinical data were collected from 981 patients from a multi-institutional international cohort with real-time polymerase chain reaction-confirmed COVID-19. Radiomics features were extracted from chest CT of the patients. The data of the cohort were randomly divided into training, validation, and test sets using a 7:1:2 ratio. A ML pipeline consisting of a model to predict severity and time-to-event model to predict progression to critical illness were trained on radiomics features and clinical variables. The receiver operating characteristic area under the curve (ROC-AUC), concordance index (C-index), and time-dependent ROC-AUC were calculated to determine model performance, which was compared with consensus CT severity scores obtained by visual interpretation by radiologists. Results: Among 981 patients with confirmed COVID-19, 274 patients developed critical illness. Radiomics features and clinical variables resulted in the best performance for the prediction of disease severity with a highest test ROC-AUC of 0.76 compared with 0.70 (0.76 vs. 0.70, p = 0.023) for visual CT severity score and clinical variables. The progression prediction model achieved a test C-index of 0.868 when it was based on the combination of CT radiomics and clinical variables compared with 0.767 when based on CT radiomics features alone (p < 0.001), 0.847 when based on clinical variables alone (p = 0.110), and 0.860 when based on the combination of visual CT severity scores and clinical variables (p = 0.549). Furthermore, the model based on the combination of CT radiomics and clinical variables achieved time-dependent ROC-AUCs of 0.897, 0.933, and 0.927 for the prediction of progression risks at 3, 5 and 7 days, respectively. Conclusion: CT radiomics features combined with clinical variables were predictive of COVID-19 severity and progression to critical illness with fairly high accuracy.

• Journal of Intelligence and Information Systems
• /
• v.20 no.4
• /
• pp.1-23
• /
• 2014

Market forecasting aims to estimate the sales volume of a product or service that is sold to consumers for a specific selling period. From the perspective of the enterprise, accurate market forecasting assists in determining the timing of new product introduction, product design, and establishing production plans and marketing strategies that enable a more efficient decision-making process. Moreover, accurate market forecasting enables governments to efficiently establish a national budget organization. This study aims to generate a market growth curve for ICT (information and communication technology) goods using past time series data; categorize products showing similar growth patterns; understand markets in the industry; and forecast the future outlook of such products. This study suggests the useful and meaningful process (or methodology) to identify the market growth pattern with quantitative growth model and data mining algorithm. The study employs the following methodology. At the first stage, past time series data are collected based on the target products or services of categorized industry. The data, such as the volume of sales and domestic consumption for a specific product or service, are collected from the relevant government ministry, the National Statistical Office, and other relevant government organizations. For collected data that may not be analyzed due to the lack of past data and the alteration of code names, data pre-processing work should be performed. At the second stage of this process, an optimal model for market forecasting should be selected. This model can be varied on the basis of the characteristics of each categorized industry. As this study is focused on the ICT industry, which has more frequent new technology appearances resulting in changes of the market structure, Logistic model, Gompertz model, and Bass model are selected. A hybrid model that combines different models can also be considered. The hybrid model considered for use in this study analyzes the size of the market potential through the Logistic and Gompertz models, and then the figures are used for the Bass model. The third stage of this process is to evaluate which model most accurately explains the data. In order to do this, the parameter should be estimated on the basis of the collected past time series data to generate the models' predictive value and calculate the root-mean squared error (RMSE). The model that shows the lowest average RMSE value for every product type is considered as the best model. At the fourth stage of this process, based on the estimated parameter value generated by the best model, a market growth pattern map is constructed with self-organizing map algorithm. A self-organizing map is learning with market pattern parameters for all products or services as input data, and the products or services are organized into an $N{\times}N$ map. The number of clusters increase from 2 to M, depending on the characteristics of the nodes on the map. The clusters are divided into zones, and the clusters with the ability to provide the most meaningful explanation are selected. Based on the final selection of clusters, the boundaries between the nodes are selected and, ultimately, the market growth pattern map is completed. The last step is to determine the final characteristics of the clusters as well as the market growth curve. The average of the market growth pattern parameters in the clusters is taken to be a representative figure. Using this figure, a growth curve is drawn for each cluster, and their characteristics are analyzed. Also, taking into consideration the product types in each cluster, their characteristics can be qualitatively generated. We expect that the process and system that this paper suggests can be used as a tool for forecasting demand in the ICT and other industries.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.13 no.6
• /
• pp.115-121
• /
• 2013

This paper related with the DSE-MMA (Dithered Sign-Error MMA) that is the simplification of computational arithmetic number in blind equalization algorithm in order to compensates the intersymbol interference which occurs the passing the nonlinear communication channel in the presence of the band limit and phase distortion. The SE-MMA algorithm has a merit of H/W implementation for the possible to reduction of computational arithmetic number using the 1 bit quantizer in stead of multiplication in the updating the equalizer tap weight. But it degradates the overall blind equalization algorithm performance by the information loss at the quantization process compare to the MMA. The DSE-MMA which implements the dithered signed-error concepts by using the dither signal before qualtization are added to MMA, then the improved SNR performance which represents the roburstness of equalization algorithm are obtained. It has a concurrently compensation capability of the amplitude and phase distortion due to intersymbol interference like as the SE-MMA and MMA algorithm. The paper uses the equalizer output signal, residual isi, MD, MSE learning curve and SER curve for the performance index of blind equalization algorithm, and the computer simulation were performed in order to compare the SE-MMA and DSE-MMA applying the same performance index. As a result of simulation, the DSE-MMA can improving the roburstness and the value of every performance index after steady state than the SE-MMA, and confirmed that the DSE-MMA has slow convergence speed which meaning the reaching the seady state from initial state of adaptive equalization filter.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.52 no.6
• /
• pp.79-84
• /
• 2015

In a distribution of digital image, there is a serious problem that is a distribution of the altered image by a forger. For the problem solution, this paper proposes a median filtering (MF) image forensic decision algorithm using a feature vector according to the pixel value's gradients. In the proposed algorithm, AR (Autoregressive) coefficients are computed from pixel value' gradients of original image then 1th~6th order coefficients to be six feature vector. And the reconstructed image is produced by the solution of Poisson's equation with the gradients. From the difference image between original and its reconstructed image, four feature vector (Average value, Max. value and the coordinate i,j of Max. value) is extracted. Subsequently, Two kinds of the feature vector combined to 10 Dim. feature vector that is used in the learning of a SVM (Support Vector Machine) classification for MF (Median Filtering) detector of the altered image. On the proposed algorithm of the median filtering detection, compare to MFR (Median Filter Residual) scheme that had the same 10 Dim. feature vectors, the performance is excellent at Unaltered, Averaging filtering ($3{\times}3$) and JPEG (QF=90) images, and less at Gaussian filtering ($3{\times}3$) image. However, in the measured performances of all items, AUC (Area Under Curve) by the sensitivity and 1-specificity is approached to 1. Thus, it is confirmed that the grade evaluation of the proposed algorithm is 'Excellent (A)'.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.20 no.1
• /
• pp.95-100
• /
• 2020

This paper related with the performance evaluation of mSE-MMA (modified Signed Error-Multi Modulus Algorithm) adaptive equalization algorithm which is possible to reduce the distortion that is occurs in nonlinear communication channel like as additive noise, intersymbol interference and fading. The SE-MMA algorithm are emerged in order to reducing the computational load compared to the presently MMA algorithm, it has the degraded equalization performance by this. In order to improve the performance degradation of SE-MMA, the mSE-MMA controls the step size according to the existence of arbitrary radius circle of equalizer output is centered at transmitted symbol point. The performance of proposed mSE-MMA algorithm were compared to present SE-MMA using the same channel and noise environment by computer simulation. For this, the recoverd signal constellation which is the output of equalizer, residual isi and MD (Maximum Distortion), MSE learning curve which is represents the convergence performance and SER which is represents the roburstness of noise were used as performance index. As a result of simulation, the mSE-MMA has more superior to the SE-MMA in every performance index, and was confirmed that mSE-MMA has roburstness to the noise in the SER performance than SE-MMA especially.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.14 no.5
• /
• pp.103-109
• /
• 2014

This paper related with the CR-MMA which is possible to improving the equalization performance by applying the concept of constellation reduction in the MMA adaptive equalization alogorithm in order to reduce the intersymbol interference that is occurred in the nonlinear communication channel. In the updating process of MMA adaptive equalizer, the error signal is being obtained by using the equalizer output, and the performance will be degraded by the increase the error signal in the high order QAM constellation. But by using the constellation reduction, the high order QAM signal will be changed to the 4-QAM signal constellation and then the error signal will be obtained. By doing so, the error signal will be minimized and it is possible to improve the equalization performance in the high order QAM transmitted signal. The Computer simulation was performed in order to compare the performance of the proposed CR-MMA algorithm and original MMA algorithm in the same communication channel and noise environment. For this, the recoverd signal constellation which is the output of equalizer, residual isi and MD (Maximum Distortion) learning curve which is represents the convergence performance and SER which is represents the roburstness of noise were used. As a result of simulation, the CR-MMA has more superior to the MMA. And it was confirmed that the CR-MMA has roburstness to the noise in the SER performance.

• Journal of KIISE:Computing Practices and Letters
• /
• v.7 no.6
• /
• pp.549-558
• /
• 2001

It is important to reuse existing motion capture data for reduction of the animation producing cost as well as efficiency of producing process. Because its motion curve has no control point, however, it is difficult to modify the captured data interactively. The motion transition is a useful method to reuse the existing motion data. It generates a seamless intermediate motion with two short motion sequences. In this paper, Uniform Posture Map (UPM) algorithm is proposed to perform the motion transition. Since the UPM is organized through quantization of various postures with an unsupervised learning algorithm, it places the output neurons with similar posture in adjacent position. Using this property, an intermediate posture of two active postures is generated; the generating posture is used as a key-frame to make an interpolating motion. The UPM algorithm needs much less computational cost, in comparison with other motion transition algorithms. It provides a control parameter; an animator could control the motion simply by adjusting the parameter. These merits of the UPM make an animator to produce the animation interactively. The UPM algorithm prevents from generating an unreal posture in learning phase. It not only makes more realistic motion curves, but also contributes to making more natural motions. The motion transition algorithm proposed in this paper could be applied to the various fields such as real time 3D games, virtual reality applications, web 3D applications, and etc.