| 亀田 友哉 | 1551036: M, 2回目発表 | コンピューティング・アーキテクチャ | 中島 康彦,池田 和司,高前田 伸也,TRAN THI HONG |
|
title: Development and Evaluation of Cellular Neural Network using Oxide Semiconductor Synapses for Character Reproduction
abstract: Recently, neural networks have been developed for variable purposes including image and voice recognitions. However, those based on software require much calculation and energy. Therefore, we are developing hardware of a cellular neural network (CNN) that has a features of scalability. In this study, we developed a CNN simulator for character reproduction. Here, each neuron is connected to only neighboring neurons by synapses, where the learning is executed by changing the connection strengths. Particularly, we assumed to use oxide semiconductors as synapses and utilize a phenomenon that the conductance changes when an electric current flows. We modeled this phenomenon and implemented it into the simulator to determine the network architecture and device parameters. This time, we confirmed that this architecture can learn four characters and deterioration rates of synapses for learning. language of the presentation: Japanese | |||
| 佐藤 敬済 | 1551048: M, 2回目発表 | ディペンダブルシステム学 | 井上 美智子,池田 和司,大下 福仁,大和 勇太 |
|
title: Accuracy improvement of LSI failure prediction using variation correction
abstract: Burn-in test is an effective test process to screen out early-life failures of VLSI circuits. However, burn-in test requires an expensive equipment and long test application time. Test data analysis provides an opportunity to reduce these costs. Burn-in test result can be predicted by machine learning using existing test result(current, voltage frequency, etc.). To accurately predict suspicious dies, it is mandatory to compensate variation effects in observation data caused by both test equipments and manufacturing process. This work addresses the issue with a variation correction method. Current progress is reported in this presentation. language of the presentation: Japanese | |||
| 嶋谷 知 | 1551051: M, 2回目発表 | コンピューティング・アーキテクチャ | 中島 康彦,井上 美智子,高前田 伸也,TRAN THI HONG |
|
title: Compact HEVC hardware encoder for light field video
abstract: In recent years, surveillance cameras are becoming high-definition in the IoT field. There are light-field format as an image format that can perform refocusing and distance image generation after the shooting, which is useful in the monitoring camera. In this research, we assume to compress and send the light-field video, and we propose a small light-field encoder for IoT. language of the presentation: Japanese | |||
| 孔 惠子 | 1551041: M, 2回目発表 | 計算システムズ生物学 | 金谷 重彦,佐藤 嘉伸,MD.ALTAF-UL-AMIN,佐藤 哲大 |
|
Title: A research on Cine-MRI image improvement with free breathing
Abstract: Cardiac MRI can evaluate heart function and diagnose ischemic cardiac disease. Cine-MRI can observe the shape and evaluate function of heart as movies by, however, there are also disadvantages of poor image quality. Generally, we can improve image quality by classifying the cardiac phase under synchronous electrocardiogram recording and performing weighted averaging the phase image. However, such approach is not applicable to the patients with arrhythmia. In this study we aim to develop a new method for classifying cardiac phases, without synchronous electrocardiogram, and improving the image quality. In this presentation I will introduce the previous related research work, and describe the proposed method. Language of the presentation: Japanese | |||
| 大谷 悠太 | 1551023: M, 2回目発表 | 生体医用画像 | 佐藤 嘉伸,金谷 重彦,大竹 義人,横田 太 |
|
title: Shape correspondence between healthy and diseased pelvis for statistical shape model abstract: In construction of Statistical Shape Model(SSM), shape correspondence between the healthy shapes (in other words, a group of shapes without pathological variation) and the diseased shapes (with pathological variation) is necessary. Correct correspondence will help SSM to represents the suitable non pathological variations together with the local severe deformation due to pathological progression. Specifically in the case of diseased hips, severe deformation due to acetabular dislocation or development of bone spurs frequently occurs. In this papar, we aim to improve shape correspondence between healthy shapes and severely diseased shapes of pelvis by manually identifying several corresponding anatomical landmarks based on the expert knowledge about the pathological progression. language of the presentation: Japanese | |||
| 山中 大幸 | 1551113: M, 2回目発表 | 生体医用画像 | 佐藤 嘉伸,金谷 重彦,大竹 義人,横田 太 |
|
title: Prediction of trabecular bone anisotropy from clinical CT image using micro CT image and statistical learning
abstract: In treatment of proximal femoral fracture, the information of bone mineral density and trabecular bone structure in femoral head is important. However, resolution of clinical CT image is not enough to analyze those information. On the other hand, the use for the living body by micro CT is limited. In the previous study, it is suggested that trabecular bone anisotropy is predicted using supervised learning and morphometric feature descriptor from a database of pairs of high resolution QCT and clinical QCT. The previous study use ex vivo images in database of both high resolution QCT and clinical QCT. In this study, the proposal method applies using in vivo clinical image to framework of the previous study. language of the presentation: Japanese 発表題目: マイクロCT画像と統計学習を用いた臨床用CT画像からの骨梁構造の異方性の予測 発表概要: 大腿骨近位部骨折における治療の際，大腿骨頭の骨密度や骨梁構造などの情報が重要になる．しかしながら，現在の臨床用CTではこれらの情報を正確に解析するのに解像度が不十分である．一方，解像度の高いマイクロCTでは生体の撮影は行えない制約がある．そこで，同一対象が撮影され位置合わせされた臨床用CT画像とマイクロCT画像が組となっているデータベースから，ランダムフォレストを拡張した学習法により臨床用CT画像から抽出した特徴量をマイクロCT画像に対応するテンソルにマッピングすることで，臨床用CT画像における骨梁構造の異方性を予測する手法が提案されている．従来法が学習データにどちらもex vivoな画像を用いているのに対して，本研究では臨床用CT画像にin vivoな画像を用いることにより，従来法を実際の臨床で得られるCT画像データに対して適用した際，どの程度の性能を発揮できるかを確かめる． | |||