小橋 昌司

肩腱板筋群の診断支援のため、人工知能を用いた肩腱板筋群のセマンティックセグメンテーション(SS)およびその領域から筋質(脂肪含有率・萎縮率)の推定を行った。本学整形外科教室より提供された被験者28名のT1強調で撮影された肩部MRのsagittal断面画像のうち、医師が指定した1枚のスライスから水と脂肪の同位相画像(in phase画像)、脂肪画像を用いた。SSのネットワークとして、現在一般的に利用されている深層学習ベースのU-netを使用した。SSの結果、肩甲下筋および棘上筋については従来の研究よりも良い精度となり、他の筋肉領域についても従来と同等の精度となった。筋質推定では、萎縮率については医師および標準的な方法に近い値が得られたが、脂肪含有率については課題が残った。T1 in phase画像のみから筋質評価を行う手法についても現在検討を行っている。

CT画像の非石領域によりESWL(extracorporeal shock wave lithotripsy)アウトカムをCNN(convolutional neural network)で予測し、その有効性を評価した。尿路結石症によりESWLを受けた175例を対象とした。CNN学習の前にデータセットを(学習:検証:テスト)=(6:2:2)の比率にランダム分割し、アウトカムのうち失敗が成功に比べて極端に少ないため、分割には層化抽出を行い、いずれのセットにも成功・失敗がおおむね等しい割合となるようにした。学習データと分割されたテストデータにおいて、AUCが0.9を超える結果が確認されたことから、非石領域はアウトカムに対して一定の予測能力を持つと考えられた。予測確率と他の因子の相関係数を確認すると、それぞれ0.16、0.01と小さく、予測確率、つまり非石領域による因子はアウトカム予測において結石の因子とは独立に、かつ高い予測能力を持つことが明らかになった。

整形外科手術は腹腟鏡手術や開腹手術と比較して手術工程および使用する手術器具が多く,外科手術中に医療器具の受け渡しを行う看護師は大きな負担を強いられている.我々は過去に人工膝関節置換術を対象とした整形外科手術における手術室看護師を支援するためのナビゲーションシステムを提案した.この研究では畳み込みニューラルネットワークを用いて手術画像全体に基づいた画像認識により手術工程の認識を試みたが,実用化に必要とされる精度には及ばなかった.本研究では整形外科手術における手術工程の認識精度の改善を実現するために,手術映像から取得したフレーム毎に物体検出(YOLO)を行い,器具のクラス情報と位置座標を検出する.スマートグラス(眼鏡型のデバイス)を用いて記録した整形外科手術映像は手術間で照明環境や撮影角度が大きく異なっており,それらの影響を低減させるための最適なデータの前処理法やデータ拡張法を検討した.(著者抄録)

In case of pediatric, about 80% intractable epilepsy patients are accompanied with cortical dysplasia. It is very effective to detect cortical dysplasia lesions in order to we plan treatment of intractable epilepsy. Although MR image analysis attracts the considerable attentions because it is non-invasive for the human body, automated methods for detecting cortical dysplasia with MR images of pediatric brain are not established yet. In addition, image features on the pediatric brain MR images of cortical dysplasia are not clarified. This paper proposes an estimation method for detecting cortical dysplasia in pediatric brain MR images. And, fractal dimension is evaluated as a new image feature of cortical dysplasia. The experimental results in four patients (6, 7, 2 and 3 years old) with cortical dysplasia showed that a mean sensitivity of 94.2 %, a mean specificity of 93.8%, and a mean efficiency of 93.8 %.

Personal classification using sole pressure change is essential for intelligent control, security on home because the users do not need to have anything, such as ID-Card, PIN code and so on. In our study, we propose a personal classification system by sole pressure change obtained by mat type pressure sensor. Mat type pressure sensor is placed on the floor in the entrance of home. We employ four features for classifying each family member and do personal classification based on Euclidean distance based method. As the experimental result on healthy 60 volunteers ranged from 20 to 80 years old, we have evaluated the performance. The results showed that the proposed system successfully classified them and it is especially useful in home intelligent system. © 2010 The Institute of Electrical Engineers of Japan.

人工膝関節置換術(Total knee arthroplasty, TKA)後の膝関節の動態解析は整形外科における研究において非常に重要な問題である．これまでの研究において2次元のX線透視画像と人工膝関節の3次元形状モデルを用いた2-D/3-D イメージマッチング手法が多く提案されているが，これらは静止画像による解析に留まり，膝関節の連続的な動作を動画像により解析する研究は少数である．また，従来の手法は高次元のパラメータ空間による局所解の問題があり，膝関節の連続的な動作は考慮されていない．本論文では，パーティクルフィルタを用いた人工膝関節の3次元動態解析システムを提案する．提案手法はパーティクルフィルタにより複数の候補を作成しパラメータの探索を行い2-D/3-D イメージマッチングを用いて人工膝関節の3次元位置姿勢を推定する．実験の結果，提案手法により膝関節の連続的な動作の解析，膝関節角度の推定が可能となった．

新生児脳疾患は脳形状の変化を伴うことが多いため，症状進行度の診断にはMR画像を用いた評価が有効である．しかし，手作業で複数のMR画像から脳表抽出を行うと医師に多大な負担が生じる．さらに，診断者間変動や診断者内変動が起こる可能性もある．そのため，新生児頭部MR画像から脳表を自動的に抽出する手法が求められている．これまでに成人頭部MR画像を対象とした脳表抽出法は多く提案されているが，新生児頭部MR画像を対象とした手法は少ない．本研究では，新生児頭部MR画像を対象とした粒子を用いる脳表抽出法を提案する．提案手法は，脳組織に見立てた3種類の粒子をMR信号値を基にして脳領域に配置し，周囲の粒子分布を基にして粒子の遷移，移動を行い，粒子の分布をMR画像と一致させることで脳表の抽出を行う．新生児頭部MR画像に対して提案手法を適用した結果，良好に脳表を抽出できた．

The brain shape is deformed regionally by kinds of cerebral diseases and the degree of progress. Therefore quantitative evaluation of the deformation using MR images is effective for diagnosis of cerebral diseases. To evaluate the cerebral deformation, almost conventional methods are based on normalization of the brain shape which deforms the evaluating brain into the standardized brain. Because the normalization process does not take into account anatomical features such as the cerebral sulci and gyri, in some cases the normalization process produces that one sulcus of the evaluating brain miss-corresponds to the other sulcus of the standardized brain. This paper proposes a homologous brain shape modeling method for quantitative evaluation of the brain shape in MR images. We define a new image feature called sulcal-distribution index (SDI) to represent the 3-D distribution of sulci, and the proposed method deforms a template brain model so that SDI of the deformed brain model calculated from the evaluating brain MR images is similar to SDI of the template brain model. By using SDI, the proposed method can take into account anatomical features of the cerebral sulci. The experimental results showed that the proposed method homologically modeled the brain shape with a mean displacement of 1.3 mm. ©2009 IEEE.

This paper proposes an ultrasonic system to determine thickness and wave speed of an object. In it, the authors newly develop a triangle probe system composed of three probes: a central probe and two side probes. The central probe straight transmits/receives ultrasonic waves and the right probe receives ultrasonic waves transmitted from the left one. In the authors' method, the triangle probe is moved to the depth direction of an object. Then, two peak intensities are revealed at the focal points of the surface and at the bottom of the object. The authors show a method to calculate the wave speed and the thickness of the objects from two focal points. In the experiment, the authors extract the two focal points aided by fuzzy logic. Then, they determine the wave speed and thickness of the object from these focal points. The experimental results on two hard phantoms show that the method can successfully determine the wave speed and the thickness of the object with high accuracy. Copyright © 2009 inderscience enterprises Ltd.

Because most of refractory patients with focal cortical dysplasia are candidate for epilepsy surgery, it is important to detect cortical dysplasia of epileptic children. This paper proposes an automated method for detecting cortical dysplasia using magnetic resonance (MR) images. To evaluate the cerebral cortex layer, the proposed method extracts texture features of a region which is perpendicular to the cerebral surface. Cortical dysplasia degree (C.D.D.) is estimated by using support vector machine (SVM). The experimental results for two subjects showed that the proposed method detected cortical dysplasia lesions with a mean sensitivity of 68.6%, a mean specificity of 67.0%, and a mean efficiency of 68.1%.

Total knee arthroplasty (TKA) is an orthopedic surgery which replaces the damaged knee joint with the artificial one. To diagnose the function of the implanted knee joint, it is effective to estimate 3-D knee kinematics in vivo. There are some conventional methods for estimating kinematics of the implanted knee using 2-D/3-D image registration for X-ray fluoroscopic images and 3-D geometrical models of the knee implant. However, these methods are based on static image analysis although the knee joint continuously moves. This paper proposes an analysis method of the knee kinematics using digital radiography images with Kalman filter. Use of Kalman filter enables us to take into account the continuous knee movement. The experimental results showed that the proposed method estimated the knee joint angles within a mean error of 0.31 deg.

This paper proposes a biometric personal authentication based on the pressure distribution while one step walking. We extract one step from a walk on a mat type load distribution sensor and use it to personal authentication. With this method, features which are based on weight movement and foot shape during walking are calculated, then a classifier is developed on the basis of fuzzy inference. We employed 30 volunteers. All volunteers are ranged from 20 to 85 years old. For each volunteer, we took walk data six times. Then, we evaluated this method by five training data and one test data. We obtained 6.1% EER (Equal Error Rate) and 13.9% FRR (False Rejection Rate) in verification (1:1 collation) and identification (1:N collation), respectively.

Cerebral surface extraction plays a fundamental role of computer aided diagnosis (CAD) for neonatal brain magnetic resonance (MR) images. However, cerebral sulci of the neonatal brains is complexity folded, and it is difficult to extract complete cerebral contour from MR images due to the limitation of spatial resolution and partial volume effect (PVE). This paper proposes a novel method to extract the cerebral contour based on fuzzy, thick rubber model (TRM). The TRM is deformed by using fuzzy control schemes so that the digitally synthesized MR images from the deforming TRM are identical to the given MR images. By synthesizing the MR images with respect to PVE, the proposed method is able to extract the cerebral contour with sub-voxel accuracy. The proposed method was applied to 7 subjects whose revised ages were from -17 days to 34 days. The root-mean-squared-error between the extracted contour and the manually delineated contour by two physicians was 1.09 +/- 0.48 mm from the truth contour. And, to demonstrate the clinical effective, gyral index was calculated using the extracted cerebral contour.

This paper describes a method for a respiratory rate monitoring system by an air pressure sensor. By using this sensor, we propose a detection method of a respiratory rate for human in bed by fuzzy logic. Our method was examined on four healthy, volunteers. We successfully detected the respiratory rate and the time of apnea state. In our method, fuzzy logic plays a primary role in the detection of respiratory points. The experimental results showed that the error ratio of respiratory rate was 1.3 % and the error of time of apnea state was 1.1 seconds. Consequently, this system can noninvasively detect the respiratory rate and the time of apnea state by using an unconstrained device.

皮質形成異常に起因する難治てんかんの外科的治療には，小児頭部MR画像からの皮質形成異常領域の検出は非常に有効である．しかし，皮質形成異常のMR画像上での画像特徴は明らかでなく，またその自動検出手法も未だ確立されていない．本文では，脳の皮質構造を評価するため，脳表に直交する評価単位領域を構築し，各領域のテクスチャ特徴量を算出し，Support Vector Machine (SVM) を用いて皮質形成異常度推定法を提案する．同提案法において，識別精度が最大となる特徴量の組み合わせを求めることで，皮質形成異常のMR画像上での画像特徴を調査する．また，本手法を適用した結果，皮質形成異常を有する小児頭部MR画像から，平均感度68％で皮質形成異常領域を検出できた．

睡眠時無呼吸症候群(SAS: Sleep Apnea Syndrome)などの睡眠障害により良質な睡眠をとれない人が近年増加している．現在，睡眠状態のモニタリングにはポリグラフ検査が一般的であるが，センサを体表につけて計測を行うため拘束性が高い．そこで本研究では非拘束に生体情報を取得する空気圧センサを用いて心拍と呼吸活動の検出を行う．空気圧センサはベッド上にいる人の生体情報を非拘束に取得することができる．提案法では取得したデータよりファジィ推論を用いて心拍・呼吸活動の検出を行う．本手法を成人男性に適用した結果，心拍は1.71%の誤差率，呼吸は2.08％の誤差率でそれぞれ検出することができた．

Measurement of cortical thickness using human brain magnetic resonance (MR) imaging can assist physicians in quantifying cerebral atrophy. Most of the conventional measurement methods assign the same class to all pixels with a similar MR signal independent of their locations, and are therefore unsuitable for MR images that have strong intensity nonuniformity (INU) artifact. We propose an automated method that locally segments the cerebral cortex using an adapted fuzzy spatial model representing the transit of MR signals from the cerebral cortex to the white matter. This method assigns fuzzy degrees belonging to brain tissues using the adaptive fuzzy spatial model for local intensity transition from the cerebral cortex to inside the cerebrum. We also introduce an evaluation method of cortex segmentation algorithms that consists of reproducibility, quantitative, and qualitative tests; we use this method to evaluate and discuss the proposed segmentation method in comparison with the conventional method.

Diagnostic imaging system is a necessity for brain diagnosis. Transcranial Ultrasonography can noninvasively image the intracranial blood flow and brain tissue in real time from only temple area of human head. However, the ultrasonic wave causes attenuation, decentration, and refraction in the skull, so the ultrasonography can not provide the transcranial brain surface image from arbitrary place. In this paper, we propose an imaging system of brain surface and skull from arbitrary places by considering the ultrasonic refraction of the skull. We do an experiment by using a cow scapula to imitate the skull bone and a biological phantom to imitate the cerebral sulcus. We first visualize the shape of scapula, and grasp the shape of scapula surface. We second remove the delay and the multi echoes of refracted wave. We third calculate the thickness of the scapula by using fuzzy inference. In the inference, we employ amplitude, correlation coefficient and the elapsed time. Finally, we calculate the refractive angle of ultrasonic wave and visualize the image referring to the refraction of ultrasonic wave. In the result of applying our method, we can estimate the thickness of scapula at all points, and successfully visualize the phantom surface image.