| 宇城 毅犠 | M, 2回目発表 | 知能コミュニケーション | 中村 哲, 安本 慶一, 田中 宏季 |
|
title: Detection of Dementia from Responses to Questions Asked by Computer Avatars
abstract: Dementia is one of the major causes of disability. It is important to detect early dementia. Some previous studies shows possibilities of detecting dementia from speech and language features. However, These studies' tasks are same interaction patterns and questions. Therefore, these studies are not suitable for daily use. This study proposes detection of dementia from responses to questions using computer avatars. The avatar asks randomly five questions from a question set which is a total of 13 questions based on consultations with neuropsychologists, and obtained answers to these questions from 24 participants (12 dementia patients and 12 non-dementia). We recorded their responses and extracted speech and language features. We classified the two groups (dementia/non-dementia) by a machine learning algorithm (support vector machines and logistic regression) using extracted features. The results showed a 0.95 detection performance in the area under the curve of the receiver operating characteristic (AUROC) and a 0.92 accuracy. This results demonstrate that our system may be able to detect early dementia. language of the presentation: Japanese | |||
| 山口 栞 | M, 2回目発表 | 知能コミュニケーション | 中村 哲, 安本 慶一, 田中 宏季 |
|
title: Efficient Constitution and Lifestyle Habits to Predict Depressive Tendency
abstract: High prevalence of depression recognizes the need to identify risk factors related to depression. This research investigated whether each gender or age group showed different associations between depressive tendency and lifestyle habits as well as constitution. The other aim was to find constitution as well as lifestyle habits that predict depressive tendency effectively. 935 participants recruited from crowdsourcing answered a questionnaire (253 questions) inquiring about their depressive tendency (6 questions), lifestyle habits, and constitution (247 questions). We calculated effect size based upon Cohen's d to elucidate the differences of relationships between depressive tendency and lifestyle habits and constitution in each gender and age group. It was found that there was not much difference of associations between them in male and female. By contrast, differences of the associations were seen among different age groups. The other analysis was classifying participants as low or high depressive tendency group automatically using their lifestyle habits and constitution as predictive features. A random forest based classification model (RF) achieved the highest accuracy, 0.97 among four models, RF, decision tree (DT), logistic regression (LR), and logistic regression with L1 regularization (LR-L1). Relatively efficient features were identified from the classification results, and included constitution, personality, and dietary and sleeping habits. language of the presentation: Japanese | |||
| 中嶋 達也 | M, 2回目発表 | インタラクティブメディア設計学 | 加藤 博一, 安本 慶一, Christian Sandor, 武富 貴史 |
|
title: Reduce of CKD Diet Theraphy Psychological Burden with Diet Support System
abstract: Medical home diets are essential methods for slowing the progression of chronic kidney disease(CKD). However, long-term practice of these therapies places an immeasurable burden on people who cook for CKD patients due to lengthy cooking procedures and precise nutrient intake. In this work, we aim to support these people by significantly reducing the time they spend on weighting ingredients. We analyzed recipes designed for CKD patients and categorized included nutrients according to their degree of error and its effect on medical diet quality. Currently, we are focusing on weighting process and its optimization through combination of digital scales and medical diet recipes. In the future, we plan to assess whether our approach leads to more satisfactory cooking experiences through an extensive user study. language of the presentation:Japanese 発表題目:CKD患者のための食事療法の長期実践の支援 発表概要: 慢性腎臓病の進行阻止のため、在宅での食事療法の長期的かつ高質な実践が重要であるが、それを妨げる多くの問題が存在する。食事療法においては調理における計量の厳密さが要求され、それに起因する調理負荷の増大も問題の一つとなっている。本研究では、電子レシピと、それに連携する計量器を用い、精度を維持しつつ計量プロセスの簡素化を図ることで、調理負荷が軽減されることを示す。これまでに、慢性腎臓病患者用の献立から栄養素分析を行った。これにより各食材の計量精度がどの栄養素にどの程度の影響を与えるかが明らかになる。現在は、慢性腎臓病用の献立における計量プロセスの分析を行い、電子レシピや、それに連携した計量器の使用にどのような工夫を行えば、そのプロセスが最適化されるかを検討している。今後は、その結果に基づき調理支援システムを構築し、実験を通して実際に調理負荷が軽減されるかどうかを検証する。 | |||
| 椛島 基嵩 | M, 2回目発表 | 生体医用画像 | 佐藤 嘉伸, 加藤 博一, 大竹 義人, Soufi Mazen |
|
title: 2D-3D Reconstruction of Distal Forearm Bones from Two-view Radiographies using Convolutional Neural Network
abstract: In the treatment of fractures, malunion is one of the important complications that may cause an impairment in ADL (Activities of Daily Living), such as pain and restricted range of motion. Corrective osteotomy is one of the common treatment option for malunion. Its surgery is obtained favorable results of corrective osteotomy based on the preoperative simulation by using the 3D bone models constructed from CT data. And applications of 3D bone model are analysis of the kinematics, biomechanical simulation, treatment planning and so on. These analyses were previously analyzed using multi-time phase CT image or dynamic radiography with single-time phase CT image. However, the necessity of acquiring CT image raises problems of increased radiation exposure and medical costs. Therefore, the purpose is constructing 3D bone model from only a few radiographies using deep learning in order to reduce radiation dose and costs. In this study, we use a network architecture based on the T-L net which combines an autoencoder and a regression network to solve the 2D-3D reconstruction problem. In this presentation, we show the result of the reconstruction of forearm bones from two-view x-ray images. language of the presentation: Japanese | |||