羽石 秀昭

The world's first open-type positron emission tomography (PET), named OpenPET, is being developed at the National Institute of Radiological Sciences, Japan. We are aiming to employ the OpenPET in radiotherapy and to track the respiratory motion of a tumor in the thoracoabdominal region of a patient. By using PET images, we expect that the tumor itself can be directly visualized without using radio-opaque markers. Our demonstration results using a small prototype OpenPET system showed that the system can output reconstructed images at about two frames per second with about a 2-s delay; this delay is mainly due to the reconstruction calculation time. In this paper, we developed a time-delay correction method for tumor tracking for the OpenPET. Since it is difficult to correct the time delay using only the tumor location at 2 s before, we assumed the introduction of another high-speed sensor to acquire the respiration phase for correction. In the proposed method, the relationship between the tumor motion and the additional sensor output signal is calculated by support vector regression and the time delay is corrected by using the obtained regression line. We simulated this time-delay correction method with computer-generated PET images which had actual respiration motions obtained from clinical magnetic resonance imaging. As a result, we could track the tumor with 1.20 +/- 0.91 mm mean error when we assumed the use of a belt-type respiratory motion sensor.

腹腔鏡下手術では、術中に血管や病変部など臓器内部の情報を取得するため、ポートから挿入可能な超音波プローブが使用される。しかし、別々のモニタに表示された腹腔鏡画像と超音波画像を同時に確認しなければならず、実空間と超音波画像の対応付けが困難である。そこで腹腔鏡画像に超音波画像をリアルタイムに重畳し、1画面の観察で画像間の対応付けを直感的に行うことが可能となるシステムを構築した。プローブ先端部は屈曲自由度を有しているため、腹腔鏡画像から屈曲角を検出する手法を用いて、腹腔鏡に対する先端部の位置姿勢情報を取得した。銅線を張ったファントムを用いたシステムの評価実験の結果、超音波画像の重畳位置精度は平均6.9pixelとなった。また腹腔鏡下で行った動物実験では、プローブ先端部の屈曲角検出率は83.1%という結果を得た。(著者抄録)

In laparoscopic surgery, to identify the location of lesions and blood vessels inside organs, an ultrasound probe which can be inserted through small incision is used. However, since surgeons must observe the laparoscopic and ultrasound images both at the same time, it is difficult to understand the correspondence between the ultrasound image and real space. Therefore, to recognize the correspondence between these two images intuitively, we developed a system for overlaying ultrasound image on the laparoscopic image. Since the tip of the probe is flexed freely, we acquired the probe tip position and orientation using a method for detecting the probe angle from laparoscopic image and information obtained from optical tracking sensor. As a result of an experiment using a wire phantom, overlaying error of ultrasound images was found to be 0.97 mm. Furthermore, the rate of probe angle detection was evaluated through an animal experiment to be 83.1%.

In interventional radiology, fluoroscopy is used to determine the position of the catheter inserted into a vessel. However, since vessels cannot be identified in fluoroscopic images, it is difficult to forward a catheter to a target region only with fluoroscopy. Thus, angiography and preoperative computed tomography (CT) images are used for the clinical purpose. CT images are useful for understanding the three-dimensional (3D) structure, but guidance of catheter is still difficult since the relationship between CT images and the fluoroscopic image is unclear. In this study, we developed a method for 3D representation of deformed vessels in CT images using an angiographic image acquired preoperatively under natural respiration and preoperative CT images. We implemented the registration algorithm and applied it to patient data. As a result, we confirmed that the vessels in CT images were correctly deformed, and a position error was two pixels in the median value.

Previously, we proposed a method for reconstructing 4D-MRI of thoracoabdominal organs that can visualize and quantify the three-dimensional dynamics of organs due to respiration. However, the data acquisition time of the method is long, say, 30 min. In this study, we assume an interleave acquisition of images with a smaller number of the encoding in the k-space to shorten the data acquisition time. We also propose to use a reconstruction technique named k-t SLR that utilizes sparse and low rank structures of the data matrix to avoid image degradation due to the small number of data acquisition. We performed a simulation experiment where we regarded 4D-MR images by our previous method as ideal images, generated down sampled data in k-space, and applied k-t SLR reconstruction to those data. We evaluated the resultant images from three viewpoints and confirmed that the combination of fast data collection with a small number of encoding and the subsequent k-t SLR reconstruction can produce high quality MR images.

The X'tal cube is a next-generation DOI detector for PET that we are developing to offer higher resolution and higher sensitivity than is available with present detectors. It is constructed from a cubic monolithic scintillation crystal and silicon photomultipliers which are coupled on various positions of the six surfaces of the cube. A laser-processing technique is applied to produce 3D optical boundaries composed of micro-cracks inside the monolithic scintillator crystal. The current configuration is based on an empirical trial of a laser-processed boundary. There is room to improve the spatial resolution by optimizing the setting of the laser-processed boundary. In fact, the laser-processing technique has high freedom in setting the parameters of the boundary such as size, pitch, and angle. Computer simulation can effectively optimize such parameters. In this study, to design optical characteristics properly for the laser-processed crystal, we developed a Monte Carlo simulator which can model arbitrary arrangements of laser-processed optical boundaries (LPBs). The optical characteristics of the LPBs were measured by use of a setup with a laser and a photodiode, and then modeled in the simulator. The accuracy of the simulator was confirmed by comparison of position histograms obtained from the simulation and from experiments with a prototype detector composed of a cubic LYSO monolithic crystal with 6 x 6 x 6 segments and multi-pixel photon counters. Furthermore, the simulator was accelerated by parallel computing with general-purpose computing on a graphics processing unit. The calculation speed was about 400 times faster than that with a CPU. © Japanese Society of Radiological Technology and Japan Society of Medical Physics 2013.

In gastrointestinal surgery, surgeon subjectively judges if the organ is healthy from the color. However it is difficult to discriminate a small difference of organ's color by visual inspection. In this paper, we focus on the tissue oxygen saturation (StO(2)) that represents balance of oxygen demand and supply in tissue and try to estimate its value by transmitted light intensity analysis. We developed a system for measurement of transmitted light intensity using a compact spectrometer and a halogen light source and collected transmitted light intensity data from pig's small intestines. Absorbance of the tissue was then calculated from those data. On the basis of Beer-Lambert law, we estimated StO(2) from the calculated absorbance. Results of evaluation experiment to pig's small intestines suggested the possibility of quantitative evaluation of tissue viability by the proposed method.

腹腔鏡下手術では，術中に血管や病変部など臓器内部の情報を取得するため，ポートから挿入可能な超音波プローブが使用される．しかし，別々のモニタに表示された腹腔鏡画像と超音波画像を同時に確認しなければならず，実空間と超音波画像の対応付けが困難である．そこで腹腔鏡画像に超音波画像をリアルタイムに重畳し，1画面の観察で画像間の対応付けを直感的に行うことが可能となるシステムを構築した．プローブ先端部は屈曲自由度を有しているため，腹腔鏡画像から屈曲角を検出する手法を用いて，腹腔鏡に対する先端部の位置姿勢情報を取得した．銅線を張ったファントムを用いたシステムの評価実験の結果，超音波画像の重畳位置精度は平均6.9 pixelとなった．また腹腔鏡下で行った動物実験では，プローブ先端部の屈曲角検出率は83.1％という結果を得た．

肝胆膵外科領域において,体内に形成された固形がんに対して高精度に病変部位を検出し,内視鏡等を用いて低侵襲かつ高精度に治療することが求められている.著者らは,このような消化器がんに対する高精度低侵襲手術を実現することを目指し,診断から治療までのプロセスに導入する各種の新規イメージング技術やデバイス技術の研究開発を進めている.具体的課題として,診断・事前データ収集のフェーズでは,各種断層像からの臓器のセグメンテーション,粘弾性特性や音響特性など臓器物性のイメージング技術,呼吸による臓器変形量の計測,治療計画のフェーズでは,体位や呼吸による臓器変形の予測・モデル化,手術工程解析,治療のフェーズでは,腹腔鏡画像の視野拡大や,腹腔鏡画像と超音波画像との融合による手術支援,ステレオX線透視像による術中利用技術,などの研究開発が挙げられる.本講演では,プロジェクトの体制とこれらの課題のいくつかを紹介する.

腹腔鏡下手術では、術中に血管や病変部など臓器内部の情報を取得するため、ポートから挿入可能な超音波プローブが使用される。しかし、別々のモニタに表示された腹腔鏡画像と超音波画像を同時に確認しなければならず、実空間と超音波画像の対応付けが困難である。そこで腹腔鏡画像に超音波画像をリアルタイムに重畳し、画像間の対応付けを直感的に行うことが可能となるシステムを構築した。プローブ先端部は屈曲自由度を有しており、手元のダイヤル操作により角度が変化する。そこで先端部の位置姿勢を取得するため、腹腔鏡-プローブ間の位置関係と腹腔鏡画像上のプローブ形状から先端部の角度を推定する手法をシステムに統合した。これにより腹腔内に挿入される先端部にセンサやマーカを装着することなく位置姿勢を取得することが可能となった。動物実験により本システムの有効性を検証した。(著者抄録)

腹腔鏡下手術では術中に血管や病変部など臓器内部の情報を取得するため,ポートから挿入可能な超音波プローブが使用される.しかし別々のモニタに表示された腹腔鏡画像と超音波画像を同時に確認しなければならず,実空間と超音波画像の対応付けが困難である.そこで腹腔鏡画像に超音波画像をリアルタイムに重畳し,画像間の対応付けを直感的に行うことが可能となるシステムが求められている.プローブ先端部は屈曲自由度を有しているため,腹腔鏡-プローブ間の位置関係と腹腔鏡画像上のプローブ形状から先端部の角度を推定する手法を提案した.評価画像を用いた精度評価の結果,屈曲角度の平均誤差は1.1度となった.

Many aspects of medical imaging require innovative measures to improve the quality of diagnosis and treatment. The FERMI Project is promoting these innovations through new developments and improvements of high-dimension, high-definition, quantification of physical or physiological parameters of biological objects and integration of multimodal medical images. The FERMI Project consists of three sub projects: (1) development of essential imaging technologies in medicine; (2) development of dynamic imaging technologies; and (3) integration of spatial information in medical treatment. In this paper, some ongoing research studies are reviewed. These include: ultrasound-based tissue characterization; magnetic resonance elastography; 4D-MRI reconstruction; knee motion analysis from X-ray fluoroscopy and CT images; projector-based assistance for laparoscopic surgery; and a surgical navigation system with three-dimensional ultrasound imaging.

Brace treatment has been selected for knee osteoarthritis (OA) patient as a conservative treatment. However, it is difficult to evaluate the effects of braces on the three-dimensional (3D) motion of the knee joint and thus there was no clear criterion for physician to choose a brace for each patient. Previously, we developed a 2D/3D registration technique using bi-plane fluoroscopic images and 3D computed tomography image to provide the 3D motion information of the knee joint and conversion procedure of 3D motion information to angular and displacement parameters which are clinically used. These parameters represent relative posture angles and relative positions from the femur to the tibia or to the patella, respectively. In this study, we applied this technique to assess and compare effect of two kinds of braces on the femorotibial and patellofemoral joints Data were collected from ten knee OA patients and analyzed. Effect of braces for 3D relative motion between femur and tibia, or femur and patella with knee OA could be assessed through the use of highly accurate 2D/3D registration technique. Effects for displacement parameters on the both joints were very small. As for the angular parameters on femorotibial joint, both braces reduced the abduction. Two angular parameters of patellofemoral joint, abduction and internal were influenced by the both braces.

We analyzed a statistical model of diaphragm motion using regular principal component analysis (PCA) and generalized N-dimensional PCA (GND-PCA). First, we generate 4D MRI of respiratory motion from 2D MRI using an intersection profile method. We then extract semiautomatically the diaphragm boundary from the 4D-MRI to get subject-specific diaphragm motion. In order to build a general statistical model of diaphragm motion, we normalize the diaphragm motion in time and spatial domains and evaluate the diaphragm motion model of 10 healthy subjects by applying regular PCA and GND-PCA. We also validate the results using the leave-one-out method. The results show that the first three principal components of regular PCA contain more than 98% of the total variation of diaphragm motion. However, validation using leave-one-out method gives up to 5.0mm mean of error for right diaphragm motion and 3.8mm mean of error for left diaphragm motion. Model analysis using GND-PCA provides about 1mm margin of error and is able to reconstruct the diaphragm model by fewer samples.

The microcirculation is defined as the smallest vessels where gas and nutrient are exchanged with tissues. In treatment of critical care, it is desired to monitor microcirculatory dysfunction to feedback it. The recent development of new imaging modalities such as Sidestream Dark Field (SDF) imaging has helped to directly investigate microcirculation for clinical and experimental studies. In this study, in order to acquire clinically useful information from more sophisticated analysis of the SDF images, we have made a trial model for SDF imaging. Four kinds of LEDs were attached to a miniature CCD camera to acquire images under proper illumination. First, we observed the blood flow in human sublingual micro-circulation and pig mucosal microcirculation. We then conducted an experiment on biological phantom to analyze micro-circulation of the oxygen saturation. In this paper, we present the principle and the specifications of the trial model as well as some images obtained in the experiments. © 2013 Springer-Verlag.

When surgeons resect and reconstruct a hollow organ in an abdominal surgery, it is necessary to judge whether the organ to be transplanted is healthy or not. Although the condition of the organ is reflected to the color and appearance, the difference is sometimes subtle. This judgment is done based on experience and a sense of surgeons. Technologies supporting the visual diagnosis should be developed. Optimally designed illuminant based on properties of human visual system may clarify the color difference of the attention area. On the other hand, since conventional operations are conducted under the white shadowless light source, it is not preferable that color of the optimal illuminant is different from that of white shadowless light. In this paper, for the purpose of clarifying the blood circulation, the optimal white illuminant is designed by combining some kinds of LEDs. The reflectance spectra of internal organs were collected from a pig's small intestine. The spectral reflectance was also calculated from images captured by a multispectral camera which have many narrowband filters. The calculated spectral reflectance of the small intestine is used for the design of the optimal LED illuminant. It was found that the optimal illuminant can clarify the blood circulation better than the conventional illuminant. © 2013 Springer-Verlag.

Previously, we proposed a 2D/3D registration method that uses Powell's algorithm to obtain 3D motion of a knee joint by 3D computed-tomography and bi-plane fluoroscopic images. The 2D/3D registration is performed consecutively and automatically for each frame of the fluoroscopic images. This method starts from the optimum parameters of the previous frame for each frame except for the first one, and it searches for the next set of optimum parameters using Powell's algorithm. However, if the flexion motion of the knee joint is fast, it is likely that Powell's algorithm will provide a mismatch because the initial parameters are far from the correct ones. In this study, we applied a hybrid optimization algorithm (HPS) combining Powell's algorithm with the Nelder-Mead simplex (NM-simplex) algorithm to overcome this problem. The performance of the HPS was compared with the separate performances of Powell's algorithm and the NM-simplex algorithm, the Quasi-Newton algorithm and hybrid optimization algorithm with the Quasi-Newton and NM-simplex algorithms with five patient data sets in terms of the root-mean-square error (RMSE), target registration error (TRE), success rate, and processing time. The RMSE, TRE, and the success rate of the HPS were better than those of the other optimization algorithms, and the processing time was similar to that of Powell's algorithm alone. © 2012 Japanese Society of Radiological Technology and Japan Society of Medical Physics.

One of the challenging applications of PET is implementing it for in-beam PET, which is an in situ monitoring method for charged particle therapy. For this purpose, we have previously proposed two geometries for our open-type PET scanners named OpenPET. The original OpenPET had a physically opened field-of-view (FOV) between two detector rings through which irradiation beams pass. This dual-ring OpenPET (DROP) had a wide axial FOV including the gap. But, this geometry is not necessarily the most efficient for application to in-beam PET in which only a limited FOV around the irradiation field is required. Therefore, we proposed a single-ring OpenPET (SROP) geometry which can provide an accessible and observable open space with higher sensitivity and a reduced number of detectors compared to DROP. Also, we proposed two arrangement types, a slanted ellipse (SE) type where oval detector rings are slanted and stacked and an axial shift (AS) type where block detectors originally forming a conventional PET scanner are axially shifted little by little. Previously, we developed the SE-SROP prototype, and showed its good performance for in-beam PET. However, the SE-SROP decreased sensitivity slightly relative to the ring-type PET and needed a large gantry in the axial direction for conventional PET studies. On the other hand, AS-SROP can transform conventional ring-type PET scanners into stacked detectors with a gradual axial shift. In this work, we developed and evaluated a small AS-SROP prototype for proof-of-concept. The AS-SROP prototype was designed with 2 cylinder-shaped detector rings of 16 DOI detectors. The system sensitivity and spatial resolution were measured and were 5.1% and 2.6 mm, respectively, for the OpenPET mode. For the conventional PET mode, the system sensitivity and spatial resolution were 7.3% and 2.2 mm, respectively. We concluded that the AS-SROP geometry has a good potential for not only in-beam imaging but also conventional PET studies. © 2013 IEEE.

A novel DOI detector with a stack of planer scintillators (SOPS) allows easier position estimation and simpler fabrication. To investigate suitable optical characteristics of the proposed detector, we constructed a Monte Carlo simulator. In this paper, we studied the relationship between the surface roughness of monolithic scintillator and the spatial resolution. We defined the parameter of scintillator's surface roughness as roughness rate which was used to statistically determine triangle patch's angle in the simulator. We simulated a six layer SOPS and found that smoother surface leads to higher spatial resolution for x and y direction. Furthermore we applied laser-processed boundaries to the SOPS (SOPS with LPB). We confirmed that SOPS with LPB provided better spatial resolution although it leads to additional process in fabrication to some extent. © 2013 IEEE.

We are developing the OpenPET which can provide an open space to make the patient observable and accessible during PET measurements. In addition, we have proposed a real-time imaging system for the OpenPET which is expected to be used in PET-guided tumor tracking radiation therapy. We conducted a feasibility study on tumor tracking using 18F-FDG by Monte Carlo simulation of a human body-sized OpenPET and showed that the tumor tracking is feasible with a time window of 0.5 s if the tumor contains sufficient radioactivity. Although we have demonstrated real-time tumor tracking using a small OpenPET prototype, the delay was 2 s. On a human body scale, the image matrix size become huge compared to the body size of the small prototype and delay can be increased. Therefore, further improvement of the reconstruction speed is essential. From this viewpoint, image reconstruction of the entire field-of-view (FOV) is time consuming and not required for the purpose of tumor tracking. In this study, we developed a region-of-interest (ROI) reconstruction method for the OpenPET and evaluated the tumor tracking accuracy. Although conventional wisdom states that the ROI reconstruction requires a priori information of a small region inside the ROI and 180° scanning without angular truncation for accurate image reconstruction, the computer simulation showed that the ROI reconstruction method without any a priori information had equivalent accuracy in terms of tumor tracking compared with the case reconstructing the entire region. The ROI reconstruction method was more than 10 times faster than the entire FOV reconstruction method. © 2013 IEEE.

We are developing the world's first, open-type 3D PET scanner "OpenPET" for PET-image guided particle therapy such as in situ dose verification and direct tumor tracking. Following our first idea of a dual-ring Open PET (DROP), we proposed our second-generation geometry, single-ring OpenPET (SROP), which is more efficient than DROP in terms of manufacturing cost and sensitivity. In this paper, we have developed a SROP prototype based on a novel detector arrangement, in which block detectors originally forming a conventional PET scanner were axially shifted little by little. Sixteen detector units each of which consists of two depth-ofinteraction detectors are arranged to form a perfect circle, 25cm in diameter. Detector units have an axial shifting mechanism so that they can be transformed into the SROP; adding this mechanism to the units allows us to use the scanner as a conventional (i.e., non-open) PET when in-beam PET measurements are not required. After confirming its basic imaging performance using a phantom filled with F-18 solution, we carried out in-beam imaging tests in the Heavy Ion Medical Accelerator in Chiba (HIMAC). In addition to the usual carbon (C-12) beam, we applied RI beams of C-11 and C-10. Stopping positions of primary particles were directly imaged with the RI beam irradiation, while the stopping position distribution of secondary particles was imaged with the C-12 beam irradiation. Phantom study results with pencil beam irradiation of about 2.5Gy showed that beam stopping positions can be measured with the precision better than 2mm with the C-11 beam irradiation followed by 20 mm PET measurement. With the C-10 beam, PET measurement time could be reduced to 1/10 while still maintaining the precision. For both C-11 and C-10, there is room for further reduction of PET measurement time.

We are developing the world's first, open-type 3D PET scanner "OpenPET" for PET-image guided particle therapy such as in situ dose verification and direct tumor tracking. Following our first idea of a dual-ring Open PET (DROP), we proposed our second-generation geometry, single-ring OpenPET (SROP), which is more efficient than DROP in terms of manufacturing cost and sensitivity. In this paper, we have developed a SROP prototype based on a novel detector arrangement, in which block detectors originally forming a conventional PET scanner were axially shifted little by little. Sixteen detector units each of which consists of two depth-ofinteraction detectors are arranged to form a perfect circle, 25cm in diameter. Detector units have an axial shifting mechanism so that they can be transformed into the SROP; adding this mechanism to the units allows us to use the scanner as a conventional (i.e., non-open) PET when in-beam PET measurements are not required. After confirming its basic imaging performance using a phantom filled with F-18 solution, we carried out in-beam imaging tests in the Heavy Ion Medical Accelerator in Chiba (HIMAC). In addition to the usual carbon (C-12) beam, we applied RI beams of C-11 and C-10. Stopping positions of primary particles were directly imaged with the RI beam irradiation, while the stopping position distribution of secondary particles was imaged with the C-12 beam irradiation. Phantom study results with pencil beam irradiation of about 2.5Gy showed that beam stopping positions can be measured with the precision better than 2mm with the C-11 beam irradiation followed by 20 mm PET measurement. With the C-10 beam, PET measurement time could be reduced to 1/10 while still maintaining the precision. For both C-11 and C-10, there is room for further reduction of PET measurement time.

We are developing the OpenPET which can provide an open space observable and accessible to the patient during PET measurements. In addition, we have proposed the real-time imaging system for the OpenPET which is expected to be used in PET-guided tumor tracking radiation therapy and demonstrated its tracking ability using a point source and a small OpenPET prototype. However, tumor tracking in the human body still remains as a challenging task when we use 18F-FDG which is the best available tracer for tumors because of its background activity, scatter and attenuation in the body. In this study, we assess conditions under which tumor tracking is feasible in the human body by using the 4D XCAT phantom which is a realistic 4D human whole body phantom. To simulate realistic 18F-FDG distributions, we assigned standardized uptake values (SUVs) to normal organs based on the literature. We conducted Monte Carlo simulation of a human-sized OpenPET geometry by using Geant4 Application for Tomographic Emission (GATE) ver. 6.1 assuming a measurement at 100 minutes after the 18F-FDG injection of 370 MBq and a spherical tumor with the diameter of 10 to 30 mm and SUV of 3 to 10. List-mode data were generated for each 0.5 s time frame of a respiratory cycle of 5 s. Image reconstruction was done in a frame-by-frame manner and tumor position was automatically extracted for each time frame by a pattern matching technique. Tumor movement in the 4D XCAT phantom was about 17 mm at the maximum. The mean error of the tumor positions extracted from the reconstructed images was similar to the PET image resolution when the tumor size was 20 mm or more and SUV was 5 or more. We showed that tumor tracking by the OpenPET is feasible even in the human body scale and for realistic conditions. © 2012 IEEE.

In positron emission tomography (PET), 3D iterative image reconstruction methods have a huge computational burden. In this paper, we developed a list-mode image reconstruction method using graphics processing units (GPUs). Efficiency of acceleration for GPU implementation largely depends on the method chosen, where a reduced number of conditional statements and a reduced memory size are required. On the other hand, accurate system models are required to improve the quality of reconstructed images. Various accurate system models for conventional CPU implementation have been proposed, but these models basically require many conditional statements and huge memory size. Therefore, we developed a new system model which matches GPU implementation better. In this model, the detector response functions, which vary depending on each line of response (LOR), are pre-computed in CPUs and modeled by sixth-order polynomial functions in order to reduce the memory size occupied in GPUs. Each element of a system matrix is obtained on-the-fly in GPUs by calculating the distance between an LOR and a voxel. Therefore the developed system model enables efficient GPU implementation of the accurate system modeling with a reduced number of conditional statements and a reduced memory size. We applied the developed method to a small OpenPET prototype, in which 4-layered depth-of-interaction (DOI) detectors were used. For image reconstruction, we used the dynamic row-action maximum likelihood algorithm (DRAMA). Compared with a conventional model for GPU implementation, in which DRFs are given as a Gaussian function of fixed width, we saw no remarkable difference for DOI data, but for non-DOI data the proposed model outperformed the conventional at the peripheral region of the field-of-view. The proposed model had almost the same calculation time as the conventional model did. For further acceleration, we tried parallel GPU implementation, and we obtained 3.8-fold acceleration by using 4 GPUs.

臨床においてはショック症状などに対し薬物治療が行われ、血流の乱れの改善、及び物質交換機能の改善が図られる。この治療効果を確認するため微小循環のモニタリングが望まれている。我々はこれらの課題に取り組むため、まずチップ型LED(Light Emitting Diode)と小型CCDカメラを組み合わせた微小循環の撮影装置を試作した。ここでは分光特性の異なる3色のチップ型LEDを配置し、異なる照明下でのバンド画像収集を可能としている。またこの装置を用いてヒトの舌下やブタ小腸の微小循環を撮影し、赤血球の流れる様子を確認した。さらに、微小循環のバンド画像を用い、画素値の大小とヘモグロビン特性との関係を利用し、血管間の酸素濃度の違いを考察した。(著者抄録)

横隔膜の呼吸性運動を主成分分析(PCA)および一般化N次元主成分分析(GND-PCA)を用いて統計解析した。まずわれわれが以前提案した交差プロファイル法によって得られた4次元MRIから横隔膜領域を抽出した。異なる形状の横隔膜の統計解析を行うために、抽出した横隔膜を水平面に射影した後、その形状を長方形に規格化した。この長方形のマトリクスには横隔膜の体軸方向座標が入り、これが1呼吸周期分存在するため、個々のデータは3次元データとなる。従来のPCAはこれを長い1次元ベクトルとして扱い、GND-PCAはこれを3次元のまま扱う。10名の健常者の呼吸性運動に対して、それぞれの手法でモデル化を行いLeave-one-out法で評価したところ、PCAの場合に平均誤差が4mmを超えるのに対して、GND-PCAの適切なパラメータ設定で平均誤差を1mm程度に抑えられることがわかった。(著者抄録)