| 児山 昂生 | 1551045: M, 2回目発表 | 生体医用画像 | 佐藤 嘉伸,加藤 博一,大竹 義人,横田 太 |
|
title: Analysis of pelvis and femur alignment in the standing position using automated 2D-3D registration: Toward large-scale population analysis abstract: Patient specific measurement of parameters of the hip alignment in standing position, such as pelvic sagittal inclination and femoral functional anteversion angle, provides useful information in surgical planning of total hip arthroplasty. For example, preoperative planning of the anteversion angle of the implant is created based on the femoral functional anteversion angle. CT image allows measurement of these parameters only in supine position, while the parameters in the standing position is also important. The purpose of this study is to analyze the bone alignment using a hip radiograph. In order to perform the analysis on a large-scale patient database, we propose a fully automatic system by combining previously proposed two algorithms: an automated bone segmentation algorithm and a robust 2D-3D (CT-to-radiograph) registration algorithm. To evaluate accuracy and robustness of the proposed method, we performed simulation studies and experiments using radiograph of a physical phantom model with nine metallic BB markers of 1.5mm diameter. In simulation study, 2D-3D registrations on a CT scan and 100 randomly generated digitally reconstructed radiographs were performed. The mean projection distance error (PDE) was 0.726 mm and the mean femoral rotation error was 0.187 degree. In the study using a physical phantom model, the mean PDE at the BB markers was 0.227 mm. In addition, we will report the results of the measurement of 150 patients using the proposed method. language of the presentation: Japanese | |||
| 堀本 悠司 | 1551097: M, 2回目発表 | 生体医用画像 | 佐藤 嘉伸,加藤 博一,大竹 義人,横田 太 |
|
title: AN AUTOMATED METHOD FOR MEASURING FEMORAL ANTEVERSION ANGLE
abstract: The femoral anteversion angle is measured in the treatment such as total hip arthroplasty and femoral fracture. The femoral anteversion angle is defined as the anterior inclination of the neck axis relative to the femoral coordinate system. We propose a method that automatically determines a sub-region of the femoral surface corresponding to the neck by using the statistical shape model and robustly computes the neck axis by computing a center line of the sub-region. First, we compute an average femur shape from a training dataset including 100 patients who underwent total hip arthroplasty. Then, we manually identify the neck region on the average shape. Given a CT of the target patient, we apply an automatic segmentation that we previously developed and register the average shape to the target femur to establish correspondence of the vertices, which provides the neck region of the target femur. Then, we compute a series of slices that naturally cut the neck region at approximately constant interval using a method called Harmonic function field. Finally, we define the neck axis as a center line of those contours. To evaluate the accuracy of the proposed method, we compared the automatic result with the gold standard measured by an expert surgeon. We tested feasibility of the proposed method in a large population study using CT image of 153 patients including hip osteoarthritis. All the patients we analysed had operation on one side and no operation the other side, which allowed us comparative studies. language of the presentation:Japanese | |||
| INSISIENGMAY ALIVANH | 1551123: M, 2回目発表 | 自然言語処理学 | 松本 裕治,中村 哲,新保 仁,進藤 裕之 |
|
title: Lao word segmentation and part of speech tagging based Deep neural networks
abstract: Lao word segmentation and part of speech tagging are basically two steps which segment a sequence of Lao characters which do not have delimiters into words and assign part of speech tags (POS tags) for each segmented words. They are considered essential steps for high-level NLP tasks such as parsing and information extraction and more importantly, there is no research doing those two tasks concurrently. Thus, proceeding this task will be very beneficial for Lao language processing research in the future. Moreover, Deep Neural Networks approaches recently has played an important role and obtained very high performance across many different NLP tasks. As a result, Deep Neural Networks is an approach which we would like to apply to perform word segmentation and part of speech tagging for Lao language. By using deep neural networks approach, we can avoid task-specific feature engineering and use deep layers of neural networks to capture relevant features to the task. Ultimately, by giving the input of plain text that consists of characters the system will output the segmented words with part of speech tag assigned. language of the presentation: English | |||
| NURUL FITHRIA LUBIS | 1551128: M, 2回目発表 | 知能コミュニケーション | 中村 哲,松本 裕治,Sakriani Sakti,Graham Neubig,吉野 幸一郎 |
|
title: Study of emotion in speech for spoken dialogue system
abstract: Emotion has been shown to help attain a more natural and enjoyable Human Computer Interaction (HCI). However, a large part of the existing research mainly focus on the recognition and simulation of emotion. In this research, I try to analyze the two-way role of emotion in human spoken dialogue and incorporate it into human computer interaction. I will explain in detail about the role of emotion in social-affective interaction, how speakers are expressing and affected by their conversational counterpart, and how we can augment existing dialogue system with such knowledge to improve the quality of human-computer interaction. language of the presentation: English | |||