M Hori

Osaka City University, Ōsaka, Ōsaka, Japan

Are you M Hori?

Claim your profile

Publications (693)3263.3 Total impact

  • [Show abstract] [Hide abstract]
    ABSTRACT: PURPOSE Differentiation between benign and malignant liver tumors at CT can be difficult with overlapping imaging features, leading to required biopsy. Our purpose was to develop a CADf scheme to improve the diagnostic accuracy and reduce “unnecessary” biopsies. METHOD AND MATERIALS Our database consisted of 39 tumors, including 15 hepatocellular carcinomas (HCCs) in 11 patients and 24 hemangiomas in 12 patients, selected to cover a wide spectrum of tumor patterns. All lesions were confirmed either pathologically or with follow-up imaging studies over 2 years. Lesion sizes ranged from 8-69 (mean: 23) mm. Scans were acquired with a multi-detector (16 or 64 rows) CT system with 2-5 mm reconstruction slice thickness. We developed CADf for determining the likelihood of malignancy. First, a geodesic active contour model with level-set algorithms segmented the liver accurately. Then, nonlinear gray-scale conversion enhanced both hyper- and hypo-dense tumors in the segmented liver. A gradient magnitude filter followed by a thinning operator determined the precise locations of the tumor boundaries. A watershed algorithm segmented tumors by using the boundary locations for feature analysis. Based on eight morphologic and texture features selected with stepwise feature selection, artificial neural network regression (ANNR) capable of operating on continuous values was trained to distinguish malignant from benign tumors. We transformed the ANNR’s output values to the likelihood of malignancy with a maximum-likelihood estimated binormal model. The performance of our ANNR was compared with linear discriminant analysis (LDA) as reference. RESULTS Our ANNR-based CADf achieved an area under the receiver operating characteristic curve of 0.96 in differentiation between HCCs and hemangiomas in a leave-one-lesion-out cross-validation test, whereas LDA achieved 0.79. The difference was statistically significant (P=.03). Our scheme provided an accurate likelihood of malignancy for “difficult” cases, e.g., a low value (5%) for an HCC-looking hemangioma. At a specific operating point, our scheme correctly characterized 100% (15/15) of HCCs and 92% (22/24) of hemangiomas. CONCLUSION Our CADf was able accurately to differentiate HCC from hemangioma at CT and provided an accurate likelihood of malignancy. CLINICAL RELEVANCE/APPLICATION CADf could be useful for differentiating malignant from benign liver tumors in CT, potentially reducing unnecessary biopsies.
    Radiological Society of North America 2010 Scientific Assembly and Annual Meeting; 11/2010
  • [Show abstract] [Hide abstract]
    ABSTRACT: PURPOSE To retrospectively investigate incidence and risk factors for development of hypervascular hepatocellular carcinoma (HCC) from hypo-intense nodules at hepatobiliary phase (HBP) of Gd-EOB-DTPA enhanced MRI (EOB-MRI). METHOD AND MATERIALS In 140 patients (pts) who underwent EOB-MRI for known or suspected HCC at least twice, 50 pts with 50 hepatic nodules (the largest nodule was selected if several nodules existed for each patient) which showed both hypo-intense at HBP and hypovascularity at arterial phase, compared to surrounding liver, were enrolled in this study. Arterial phase and hepatobiliary phase (at 20 min post injection) were obtained after bolus EOB injection (0.1mL/kg, a rate of 2mL/sec) by 1.5 T or 3T MRI at 61 - 477 days (mean 190 days) as interval date. 50 hypo-intense nodules at HBP in 1st EOB-MR examination (baseline) were followed, and outcome was analyzed. Incidence rate of development of hypervascular HCC from hypo-intense nodule at HBP was investigated. We employed the criteria as cut-off size (group A ≥ 15 mm, B < 15mm) of hypo-intense nodule at baseline and growth size (group C ≥ 0.015 mm/day, D < 0.015 mm/day) of hypo-intense nodule compared to previous MR. Incidence of developing hypervascular HCC from hypo-intense nodule was compared between group A and B, group C and D. RESULTS Hypervascular HCCs were confirmed in 5 pts by EOB-MR, 6 pts by dynamic CT and 1 pt by CTHA, and the incidence rate of change from hypo-intense nodule to hypervascular HCC was 24% (12 of 50 pts). Intervals between MR examinations in group A (189+/-104 days) and B (200+/-120 days), or C (185+/-110 days) and D (185+/-114 days) were not significantly different (P=0.94, 0.98, respectively, Mann-Whitney’s test). Incidence rate from hypo-intense nodule to hypervascular HCC in group A (3 of 9 pts; 33%) was higher than that (9 of 41 pts; 22%) of group B (P=0.5), and that of group C (8 of 26 examinations; 31%) was significantly higher than that of group D (4 of 54 examinations; 7%) (P=0.006). CONCLUSION The incidence rate of developing hypervascular HCC from hypo-intense nodule was 24%. We should pay attention in growth size ≥ 0.015 mm/day of hypo-intense nodule at HBP compared to previous EOB-MRI. CLINICAL RELEVANCE/APPLICATION EOB-MRI plays a major role in surveillance for chronic liver disease, and incidence of developing hypervascular HCC from hypo-intense nodules at HBP is 24% (mean interval of 190 days of follow-up MR).
    Radiological Society of North America 2010 Scientific Assembly and Annual Meeting; 11/2010
  • [Show abstract] [Hide abstract]
    ABSTRACT: To compare a three-dimensional (3D) T2-weighted turbo spin-echo (TSE) magnetic resonance (MR) sequence (VISTA; Philips Medical Systems, Best, the Netherlands) with a two-dimensional (2D) T2-weighted TSE sequence in terms of image quality, signal intensity (SI) difference ratios, conspicuity, and staging of uterine tumors. This retrospective study was approved by the institutional review board, and informed consent was waived. Sixty-one women (mean age, 53.0 years ± 13.7 [standard deviation]; range, 30-87 years) with cervical carcinoma (n = 28), endometrial carcinoma (n = 21), or leiomyoma (n = 30) of the uterus were included. Patients underwent T2-weighted MR imaging at 3 T with both 1-mm-thick 3D and 5-mm-thick 2D T2-weighted TSE sequences. Three-dimensional T2-weighted TSE images were reconstructed at 5-mm thickness with the aid of a workstation. Quantitative analyses of signal-to-noise ratio (SNR) and SI difference ratios between tumors and other tissues and qualitative analyses of image quality and tissue conspicuity were performed. Two radiologists independently assessed local-regional staging for carcinomas. Quantitative values, qualitative scores, and tumor staging were analyzed by using the paired t test, Wilcoxon signed rank test, and McNemar test, respectively. Mean myometrial SNR was higher on 3D than 2D images (14.3 vs 9.8; P < .0001). Mean SI difference ratios between cervical (0.45 vs 0.34; P < .0001) or endometrial (0.46 vs 0.40; P = .044) carcinomas and gluteal muscle were higher on 3D images, but those between leiomyoma and myometrium (0.33 vs 0.43; P < .0001) were lower than those on 2D images. Image quality (P = .0004) and carcinoma conspicuity (P < .0005) were superior with the 3D T2-weighted TSE sequence. Although multiplanar reconstruction of 3D T2-weighted TSE images was useful for staging in one case, there were no significant differences between 3D and 2D T2-weighted TSE imaging in accuracy of staging for the two readers for cervical or endometrial carcinoma. The 3D T2-weighted TSE sequence showed certain advantages over the 2D T2-weighted TSE sequence, and it has the potential to improve the performance of MR imaging for the evaluation of uterine carcinoma.
    Radiology 11/2010; 258(1):154-63. · 6.34 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: To retrospectively investigate enhancement patterns of hepatocellular carcinoma (HCC) and dysplastic nodule (DN) in the hepatobiliary phase of gadolinium-ethoxybenzyl-diethylenetriamine (Gd-EOB-DTPA)-enhanced MRI in relation to histological grading and portal blood flow. Sixty-nine consecutive patients with 83 histologically proven HCCs and DNs were studied. To assess Gd-EOB-DTPA uptake, we calculated the EOB enhancement ratio, which is the ratio of the relative intensity of tumorous lesion to surrounding nontumorous area on hepatobiliary phase images (post-contrast EOB ratio) to that on unenhanced images (pre-contrast EOB ratio). Portal blood flow was evaluated by CT during arterial portography. Post-contrast EOB ratios significantly decreased as the degree of differentiation declined in DNs (1.00 ± 0.14) and well, moderately and poorly differentiated HCCs (0.79 ± 0.19, 0.60 ± 0.27, 0.49 ± 0.10 respectively). Gd-EOB-DTPA uptake, assessed by EOB enhancement ratios, deceased slightly in DNs and still more in HCCs, while there was no statistical difference in the decrease between different histological grades of HCC. Reductions in portal blood flow were observed less frequently than decreases in Gd-EOB-DTPA uptake in DNs and well-differentiated HCCs. Reduced Gd-EOB-DTPA uptake might be an early event of hepatocarcinogenesis, preceding portal blood flow reduction. The hepatobiliary phase of Gd-EOB-DTPA-enhanced MRI may help estimate histological grading, although difficulties exist in differentiating HCCs from DNs.
    European Radiology 10/2010; 20(10):2405-13. · 4.34 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: To compare enhancenent patterns of hepatocellular carcinoma (HCC) and dysplastic nodule (DN) between gadoxetate- and ferucarbotran-enhanced MRI. Patients recruited from ultrasound surveillance for HCC in chronic liver diseases were enrolled in this prospective study approved by institutional review board. Thirty-six patients with 37 histologically proven HCC, including 22 well-differentiated HCCs (wHCC), 15 moderately to poorly differentiated HCCs (mpHCCs), and 4 DNs, underwent gadoxetate-enhanced and ferucarbotran-enhanced MRI. We compared hepatobiliary phase image of gadoxetate-enhanced MRI with ferucarbotran-enhanced MR image regarding signal intensity of HCC and DN relative to surrounding liver parenchyma. We calculated contrast ratios between tumor and liver on pre-enhancement, hepatobiliary phase of gadoxetate-enhanced MRI and ferucarbotran-enhanced MRI. On ferucarbotran-enhanced MRI, all mpHCCs showed hyper-intensity, while 14 wHCCs (14/22;63%) showed iso-intensity. On hepatobiliary phase of gadoxetate-enhanced MRI, 13 mpHCCs (13/15;86%) and 20 wHCCs (20/22;91%) showed hypo-intensity. Two DNs and the other two showed iso- and hypo-intensity, respectively, on gadoxetate-enhanced MRI, whereas all DNs revealed iso-intensity on ferucarbotran-enhanced MRI. Gadoxetate-postcontrast ratio was significantly lower than ferucarbotran-postcontrast ratio in wHCC (P = 0.015). The uptake function of hepatocytes that are targeted by gadoxetate is more sensitive than that of Kupffer cells targeted by ferucarbotran in stepwise hepatocarcinogenesis.
    Journal of Magnetic Resonance Imaging 10/2010; 32(4):903-13. · 2.57 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: A computational framework is presented, based on statistical shape modelling, for construction of race-specific organ models for internal radionuclide dosimetry and other nuclear-medicine applications. This approach was applied to the construction of a Japanese liver phantom, using the liver of the digital Zubal phantom as the template and 35 liver computed tomography (CT) scans of male Japanese individuals as a training set. The first step was the automated object-space registration (to align all the liver surfaces in one orientation), using a coherent-point-drift maximum-likelihood alignment algorithm, of each CT scan-derived manually contoured liver surface and the template Zubal liver phantom. Six landmark points, corresponding to the intersection of the contours of the maximum-area sagittal, transaxial and coronal liver sections were employed to perform the above task. To find correspondence points in livers (i.e. 2000 points for each liver), each liver surface was transformed into a mesh, was mapped for the parameter space of a sphere (parameterisation), yielding spherical harmonics (SPHARMs) shape descriptors. The resulting spherical transforms were then registered by minimising the root-mean-square distance among the SPHARMs coefficients. A mean shape (i.e. liver) and its dispersion (i.e. covariance matrix) were next calculated and analysed by principal components. Leave-one-out-tests using 5-35 principal components (or modes) demonstrated the fidelity of the foregoing statistical analysis. Finally, a voxelisation algorithm and a point-based registration is utilised to convert the SPHARM surfaces into its corresponding voxelised and adjusted the Zubal phantom data, respectively. The proposed technique used to create the race-specific statistical phantom maintains anatomic realism and provides the statistical parameters for application to radionuclide dosimetry.
    Radiation Protection Dosimetry 09/2010; 141(2):140-8. · 0.91 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Catheterization is the treatment insert tube which name is catheter from thigh's or wrist's artery and discharge fluid, give medicine to affected area directly etc. Recently this treatment is commonly performed because make only a small hole to the patient body thus patient bears little risks. At the catheterization, expert surgeon's experience and knowledge are needed for making optimal shape of guidewire. There is not enough established objective criterion to decide the shape of guidewire. Therefore this research's goal is to develop semi-automated system for the determination of the optimal shape of guidewire's tip at the vessel bifurcation. In the previous research, we determined the optimal shape of guidewire two-dimensionally to reduce computation time. However, two-dimensionally determination has problem which is not shown real movement of the treatment. In this research, we extend the previous method to determine the optimal shape three-dimensionally and simulate more realistic movement of guidewire. In order to reduce the computation time, we consider the real operation of catheterization and the position of the tip of the catheter in the vessel. We decided position and direction of tip of catheter manually in this research. Moreover, we perform the phantom experiment to validate the proposed method whether the guidewire could reach the target branch vessel using the estimated shape of the guidewire. At the phantom experiment and 3D determination by proposed method, we got same result and it conforms to the real. One of the shapes of the guidewire's tip which is "A" arrive to target branch thus the other shape which is "B" doesn't arrive to target branch.
    Software Engineering and Data Mining (SEDM), 2010 2nd International Conference on; 07/2010
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Representation and evaluation methods for statistically predicting organ shapes from neighboring organ shapes are described. In order to fully utilize the constraints on interrelations of multiple organ shapes, various extents of sub-shapes of organs are considered based on their proximity instead of just using the whole organ shapes. The prediction power are evaluated for various extents of sub-shapes and enhanced by their integration. Experimental results using shape datasets of several organs in the abdominal domain demonstrate the effective of the proposed methods.
    Software Engineering and Data Mining (SEDM), 2010 2nd International Conference on; 07/2010
  • [Show abstract] [Hide abstract]
    ABSTRACT: To investigate the mechanism of enhancement of hepatocellular carcinoma (HCC) on gadoxetic acid-enhanced hepatobiliary phase magnetic resonance (MR) images and to characterize HCC thus enhanced. This retrospective study was approved by the institutional review board, and patient informed consent for research use of the resected specimen was obtained. MR images in 25 patients (20 men, five women; mean age, 68 years; range, 49-82 years) with 27 resected hypervascular HCCs (one well, 13 moderately, 13 poorly differentiated) that demonstrated hepatocyte-selective enhancement on gadoxetic acid-enhanced MR images, were quantitatively studied, and findings were correlated with results of immunohistochemical staining for a sinusoidal transporter, organic anion transporting polypeptide (OATP) 1B1 (OATP1B1) and/or OATP1B3 (OATP1B1 and/or -1B3), and a canalicular transporter, multidrug resistance-associated protein 2 (MRP2), and also with bile accumulation in tumors. Statistical analysis was performed with the Student t test and Scheffé post hoc test. Combined with positive OATP1B1 and/or -1B3 expression (O+), two patterns of MRP2 expression contributed to high enhancement: decreased expression (M-, n = 3) and increased expression at the luminal membrane of pseudoglands (M+[P], n = 3). Nodules without OATP1B1 and/or -1B3 expression (O-, n = 13) and nodules with O+ associated with increased MRP2 expression only at the canaliculi (M+[C], n = 8) induced significantly lower enhancement than those with the two expression patterns described before (O+/M- group vs O- group, P = .002; O+/M- group vs O+/M+[C] group, P = .047; O+/M+[P] group vs O- group, P < .001; O+/M+[P] group vs O+/M+[C] group, P < .001). Nodules with bile pigment (n = 12) showed significantly higher enhancement (P = .004); all five nodules (one well differentiated HCC, four moderately differentiated HCCs), which were enhanced more than adjacent liver parenchyma, contained bile pigment. High hepatocyte-selective enhancement is induced by expression patterns of transporters, which may result in accumulation of gadoxetic acid in cytoplasm of tumor cells or in lumina of pseudoglands. An HCC with gadoxetic acid enhancement is characterized by bile accumulation in tumors.
    Radiology 06/2010; 255(3):824-33. · 6.34 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Computerized liver extraction from hepatic CT images is challenging because the liver often abuts other organs of a similar density. The purpose of this study was to develop a computer-aided measurement of liver volumes in hepatic CT. The authors developed a computerized liver extraction scheme based on geodesic active contour segmentation coupled with level-set contour evolution. First, an anisotropic diffusion filter was applied to portal-venous-phase CT images for noise reduction while preserving the liver structure, followed by a scale-specific gradient magnitude filter to enhance the liver boundaries. Then, a nonlinear grayscale converter enhanced the contrast of the liver parenchyma. By using the liver-parenchyma-enhanced image as a speed function, a fast-marching level-set algorithm generated an initial contour that roughly estimated the liver shape. A geodesic active contour segmentation algorithm coupled with level-set contour evolution refined the initial contour to define the liver boundaries more precisely. The liver volume was then calculated using these refined boundaries. Hepatic CT scans of 15 prospective liver donors were obtained under a liver transplant protocol with a multidetector CT system. The liver volumes extracted by the computerized scheme were compared to those traced manually by a radiologist, used as "gold standard." The mean liver volume obtained with our scheme was 1504 cc, whereas the mean gold standard manual volume was 1457 cc, resulting in a mean absolute difference of 105 cc (7.2%). The computer-estimated liver volumetrics agreed excellently with the gold-standard manual volumetrics (intraclass correlation coefficient was 0.95) with no statistically significant difference (F = 0.77; p(F < or = f) = 0.32). The average accuracy, sensitivity, specificity, and percent volume error were 98.4%, 91.1%, 99.1%, and 7.2%, respectively. Computerized CT liver volumetry would require substantially less completion time (compared to an average of 39 min per case by manual segmentation). The computerized liver extraction scheme provides an efficient and accurate way of measuring liver volumes in CT.
    Medical Physics 05/2010; 37(5):2159-66. · 2.91 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Kupffer cell imaging is a powerful tool for the detection of liver cancer. This diagnostic procedure depends on the faculty of the reticuloendothelial system (RES) which takes up foreign bodies, including small particles. The current study aimed to develop a novel RES targeting liposomal contrast agent that functionalized with serine or mannose, the moiety specifically binding to a corresponding receptor on phagocytic cells. Liposomes loaded with non-ionic X-ray contrast media, Iohexol, were prepared by supercritical carbon dioxide reverse-phase evaporation method and were intravenously injected to healthy rabbits in order to evaluate the liver parenchymal enhancement in X-ray computed tomography (CT). From 10 to 40 min after injection, the mean enhancement value of the liver parenchyma approached 45 and 34 Hounsfield units (HU) when serine-modified iodinated liposomal contrast agent (ILCA) and mannose-modified ILCA were applied, respectively. The tumor-to-liver contrast values were also evaluated after the administration of the prepared ILCA to rabbits with VX-2 carcinoma. For serine-modified ILCA, tumor-to-liver contrast was 82 HU at 1 min and >24 HU at 10-40 min; for mannose-modified ILCA, the values were 58 HU at 0.5 min and >21 HU at 10-40 min. These vales estimated from the region of intrest and the imaging figures of liver indicate the potential of ILCA for clinical use.
    Contrast Media & Molecular Imaging 03/2010; 5(3):140-6. · 2.87 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Automatic liver segmentation on CT images is challenging because the liver often abuts other organs of a similar density. Our purpose was to develop an accurate automated liver segmentation scheme for measuring liver volumes. We developed an automated volumetry scheme for the liver in CT based on a 5 step schema. First, an anisotropic smoothing filter was applied to portal-venous phase CT images to remove noise while preserving the liver structure, followed by an edge enhancer to enhance the liver boundary. By using the boundary-enhanced image as a speed function, a fastmarching algorithm generated an initial surface that roughly estimated the liver shape. A geodesic-active-contour segmentation algorithm coupled with level-set contour-evolution refined the initial surface so as to more precisely fit the liver boundary. The liver volume was calculated based on the refined liver surface. Hepatic CT scans of eighteen prospective liver donors were obtained under a liver transplant protocol with a multi-detector CT system. Automated liver volumes obtained were compared with those manually traced by a radiologist, used as "gold standard." The mean liver volume obtained with our scheme was 1,520 cc, whereas the mean manual volume was 1,486 cc, with the mean absolute difference of 104 cc (7.0%). CT liver volumetrics based on an automated scheme agreed excellently with "goldstandard" manual volumetrics (intra-class correlation coefficient was 0.95) with no statistically significant difference (p(F<=f)=0.32), and required substantially less completion time. Our automated scheme provides an efficient and accurate way of measuring liver volumes.
    Proc SPIE 03/2010;
  • [Show abstract] [Hide abstract]
    ABSTRACT: PURPOSE Our purpose was to evaluate the usefulness of CADe of polyps in improving the sensitivity of expert radiologists in the detection of “difficult” polyps which reporting radiologists initially “missed” in CTC in a multicenter clinical trial or rated “difficult” in our retrospective review. METHOD AND MATERIALS We developed an advanced CADe system consisting of colon segmentation based on anatomy-based extraction and colon-based analysis, detection of polyp candidates based on morphologic analysis of the segmented colon, and a mixture of expert 3D massive-training artificial neural networks for reduction of multiple types of false positives (FPs). We applied our CADe scheme to 30 CTC cases, of which 20 cases had 23 polyps ≥ 6mm (mean: 10 mm) including 14 “missed” and 7 “difficult” polyps and 10 were negative cases. Four board-certified abdominal radiologists including 3 CTC experts participated in our observer performance study. They read 60 supine and prone CTC studies of the 30 patients on a CTC workstation. They were asked, first without and then with CADe, to indicate the location of polyps and their confidence level regarding the presence of the polyps. They were free to use 3D endoluminal or 2D multiplanar views for polyp detection and problem solving. They were blinded to the prevalence of polyps, but told the general performance of our CADe. The diagnostic performance was evaluated by using jackknife alternative free-response receiver-operating characteristic (JAFROC) analysis. RESULTS Our CADe achieved a sensitivity of 74% (17/23) with 3.1 (93/30) FPs per patient. With CADe, the diagnostic performance of radiologists in terms of the average figure of merit (i.e., area under the ROC curve) in the JAFROC analysis was statistically significantly improved from 0.79 to 0.83 (P=.006). The positive predictive value was also improved from 57% to 67% with CADe (P=.098), while the average by-polyp sensitivity of radiologists was improved by 10% at a statistically significant level (P=.037). The average reading time without and with CADe was 12 and 2 min per case, respectively. CONCLUSION Our CADe improved the diagnostic performance of radiologists, including expert readers, and had a great impact in detection of difficult and initially-missed polyps in CTC. CLINICAL RELEVANCE/APPLICATION Observer performance study demonstrated that CADe was effective in improving expert radiologists’ performance in detection of difficult polyps in CTC.
    Proceedings of RSNA Annual Meeting, Chicago, IL; 01/2010
  • [Show abstract] [Hide abstract]
    ABSTRACT: PURPOSE Liver cancer is the 3rd leading cause of cancer death worldwide. Early detection of hepatocellular carcinoma (HCC), the most common primary malignant tumor of the liver, can be difficult especially when attenuation difference between HCC and normal parenchyma is small. Our purpose was to develop a computer-aided diagnostic (CADe) scheme for early detection of HCC in contrast-enhanced CT. METHOD AND MATERIALS Our CADe scheme consisted of segmentation of the liver, detection of tumor candidates, feature analysis, and classification of the candidates into HCC or non-HCC lesions. Liver segmentation was performed by using a 3D geodesic-active-contour model coupled with a level-set algorithm. A nonlinear gray-scale conversion filter enhanced both hyper- and hypo-dense tumors in the segmented liver. A gradient magnitude filter was then applied, followed by image thresholding and thinning, in order to obtain precise tumor boundaries. A watershed algorithm was employed for segmenting closed-contoured regions as HCC candidates. Eight morphologic and texture features were extracted from the segmented candidates. We used our linear-output artificial neural network (LOANN) for final classification based on these features. We compared its performance with linear discriminant analysis (LDA). Our database consisted of arterial-phase hepatic CT scans of 28 patients acquired with a multi-detector-row CT system with a 16, 40, or 64 channel detector (Brilliance, Philips Medical Systems, Netherlands). Reconstructed CT slices used for CADe were 512x512 pixels in size with 3-5 mm slice thickness. Among 28 cases, 15 HCCs were found in 10 patients. All HCCs were confirmed pathologically. HCC sizes ranged from 15-43 mm with a mean of 22 mm. RESULTS The initial CADe scheme before the classification step detected 100% (15/15) of HCCs with 12.2 (342/28) false positives (FPs) per patient. Our LOANN removed 36% (124/342) of the FPs without any loss of true positive in a leave-one-out cross-validation test; thus, it yielded 100% (15/15) sensitivity with 7.8 (218/28) FPs per patient, whereas LDA yielded 80% (12/15) sensitivity at the same FP rate. CONCLUSION Our CADe scheme achieved 100% sensitivity for detection of HCCs in contrast-enhanced hepatic CT with a reasonable number of FPs. CLINICAL RELEVANCE/APPLICATION CAD could be useful for detecting HCCs in CT; thus, it would potentially improve radiologists’ sensitivity for HCCs.
    Radiological Society of North America 2009 Scientific Assembly and Annual Meeting; 11/2009
  • [Show abstract] [Hide abstract]
    ABSTRACT: To retrospectively determine the additional value of diffusion-weighted magnetic resonance imaging (MRI) to T2-weighted imaging in the evaluation of anal fistulae in comparison with gadolinium (Gd)-enhanced imaging. Thirteen patients (mean age, 35.2 years) with 20 anal fistulae were included. The protocol consisted of fat-suppressed T2-weighted fast spin-echo, diffusion-weighted single-shot echo-planar (b factors 0 and 800 s/mm(2)), and fat-suppressed Gd-enhanced T1-weighted gradient echo sequences. Two radiologists evaluated images in consensus. Eighteen (90%) fistulae were detected on T2-weighted images, and 19 (95%) and 19 (95%) were detected on diffusion-weighted and T2-weighted images combined and on Gd-enhanced and T2-weighted images combined, respectively. There was no statistically significant difference in sensitivity of the techniques (P > 0.5 for all comparison pairs). Confidence scores with diffusion-weighted and T2-weighted images combined or those with Gd-enhanced and T2-weighted images combined were significantly greater than those with T2-weighted images alone (P = 0.0047 and 0.014, respectively). Diffusion-weighted MRI of anal fistulae is a useful sequence and can be a helpful adjunct to T2-weighted imaging, especially in patients with risk factors for contrast agents.
    Journal of Magnetic Resonance Imaging 11/2009; 30(5):1021-6. · 2.57 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Segmentation of liver in CT images is regarded as a challenge in image processing due to low-contrast of datasets, variety of liver shape, and its non-uniform texture; especially for abnormal cases. In this paper, we deal with normal and abnormal datasets as images containing two or more Gaussian components. We threshold a slice in a narrow band of each mode, find liver pixels based on a priori knowledge, prepare a probability map, and threshold the map to find initial liver border. Final boundary of liver is obtained through a few iterations of `Geodesic Active Contour'. The proposed method was tested on 30 normal and 17 abnormal datasets each containing 159-263 slices; acquired from different CT machines. The results for normal and abnormal datasets are completely acceptable, according to the evaluation done by a specialist. However, for severely abnormal datasets, the proposed method is regarded as a promising algorithm for liver segmentation.
    Intelligent Information Hiding and Multimedia Signal Processing, 2009. IIH-MSP '09. Fifth International Conference on; 10/2009
  • Tonsok Kim, Masatoshi Hori, Hiromitsu Onishi
    [Show abstract] [Hide abstract]
    ABSTRACT: Central or eccentric scar has been described in (most frequently) focal nodular hyperplasia (FNH), large hemangioma, and fibrolamellar hepatocellular carcinoma, and moreover, a central scar may be rarely seen in conventional nonfibrolamellar hepatocellular carcinoma, (peripheral type) cholangiocarcinoma, and some hepatic metastases. Among the liver tumors with central or eccentric scar, differentiating FNH from other solid hepatic tumors is important because FNH usually requires no treatment. Multiphasic computed tomography and magnetic resonance imaging are useful for the differential diagnosis of liver tumors with central or eccentric scar. Liver-specific magnetic resonance contrast media will improve the differentiation between FNH and other solid hepatic tumors.
    Seminars in Ultrasound CT and MRI 10/2009; 30(5):418-25. · 1.29 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Liver cancer is one of the major death factors in the world. Transplantation and tumor removal are two main therapies in common clinical practice. Both tasks need image assisted planning and quantitative evaluations. Automatic liver segmentation is required for corresponding quantitative evaluations. Conventional approaches in liver segmentation consist of finding the initial liver border followed by tuning the border to the final mask. Finding the liver initial border is of great importance as the latter step largely depends on the initial step. In the previous works, the liver initial border was determined by applying thresholding and morphological filters. In order to estimate the liver initial boundary, we have proposed a technique based on anatomical knowledge of liver, its surrounding tissues as well as the approach that a clinician follows in screening liver in a CT dataset. Based on the above reasoning, we developed a multi-step heuristic technique to segment liver from other tissues in multi-slice CT images. The proposed technique can deal with various shapes, locations, and liver sizes. The method was evaluated in the presence of 50 actual liver data sets and the results were encouraging.
    Computerized medical imaging and graphics: the official journal of the Computerized Medical Imaging Society 09/2009; 33(8):567-87. · 1.04 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: To prospectively evaluate the image quality and diagnostic accuracy of magnetic resonance cholangiopancreatography (MRCP) at 3.0 T compared with that at 1.5 T used for patients. This study was approved by our institutional review board, and informed consent was obtained from all patients. Three-dimensional MRCP with both a 1.5-T system and a 3.0-T system was administered to 33 consecutive patients suspected of having biliary and/or pancreatic diseases. For 3-dimensional MRCP imaging, a respiratory-triggered Fast Recovery Fast Spin Echo sequence with the parallel imaging technique was used for both systems. The spatial resolution at 1.5 T was matched to that at 3.0 T, and matrix size was 512 x 160, field of view (FOV) 36 x 36 cm, slice thickness 2 mm, and the number of slices ranged from 44 to 54. Repetition time varied according to the patients' respiratory cycles, which ranged from 3000 milliseconds to 6000 milliseconds, and effective echo time, ranging from 391 milliseconds to 482 milliseconds, was automatically determined by the systems. Contrast of imaging of the common bile duct versus that of the periductal tissue was quantitatively evaluated. The image quality for homogeneity of signal intensity, image noise, susceptibility artifacts, and overall imaging quality and diagnostic accuracy for stenoses of bile and main pancreatic ducts and for pancreatic cystic lesions were qualitatively evaluated. The results for 1.5- and 3.0-T were then compared statistically, by using the paired t test for quantitative evaluation of contrast between the common bile duct and the periductal tissue, the Wilcoxon signed rank test for visual evaluation of the image quality, and the McNemar chi2 test for evaluation of sensitivity and specificity for diagnostic purposes. A P value of less than 0.05 was considered to indicate a statistically significant difference. Contrast of the common bile duct versus that of the periductal tissue at 3.0 T was superior to that at 1.5 T (P < 0.001). In the visual evaluation, image quality at 3.0 T was superior to that at 1.5 T for evaluation of image noise (P < 0.001), overall image quality (P < 0.001), and delineation of intrahepatic bile ducts (P < 0.01) and the main pancreatic duct (P < 0.05), whereas there were no significant differences in homogeneity of signal intensity and in susceptibility artifacts. The diagnostic accuracy for 3.0 T was approximately equivalent to that for 1.5 T. Our results for patients with biliary and pancreatic diseases demonstrate a marked improvement in contrast of the common bile duct versus that of the periductal tissue and image quality including image noise at 3.0 T compared by 1.5 T when identical spatial resolution configuration were used. The results suggest that 3.0 T may allow higher spatial resolution and offer promise for improved diagnosis in MRCP, although further investigations using optimize scan parameters will be needed before its full potential can be achieved.
    Investigative radiology 09/2009; 44(9):559-65. · 4.85 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: To prospectively compare magnetic resonance imaging (MRI) at 3.0 T and 1.5 T in the same patients for preoperative evaluation of endometrial carcinoma. Thirty consecutive patients with endometrial carcinoma underwent MRI at both 3.0 T and 1.5 T as well as surgery. Quantitative and qualitative analyses were performed. Two radiologists independently evaluated images. MR findings were compared with surgicopathologic findings. Image homogeneity of T2-weighted images at 3.0 T was significantly inferior to that at 1.5 T (P = 0.007). The scores of image homogeneity and susceptibility artifacts were not significantly different between 3.0 T gadolinium-enhanced imaging and 1.5 T imaging (P = 0.09 and 0.36). Kappa statistics showed good interobserver agreement between the two radiologists for local-regional staging on T2-weighted images (kappa>0.6). The area under the receiver operating characteristic curve (Az) values for T2-weighted imaging in terms of myometrial invasion, cervical invasion, and lymph node metastases were 0.88 (3.0 T) versus 0.91 (1.5 T), 0.84 versus 0.83, and 0.94 versus 0.95 for reader 1, respectively. There were no significant differences between imaging at 3.0 T and at 1.5 T in Az values for either reader (P > 0.35). 3.0 T MRI is an equivalent imaging modality to 1.5 T imaging for presurgical evaluation of endometrial carcinoma, although not significantly superior to 1.5 T imaging.
    Journal of Magnetic Resonance Imaging 09/2009; 30(3):621-30. · 2.57 Impact Factor

Publication Stats

17k Citations
3,263.30 Total Impact Points

Institutions

  • 1990–2014
    • Osaka City University
      • • Department of Radiology
      • • Graduate School of Medicine
      • • Department of Cardiology
      Ōsaka, Ōsaka, Japan
  • 2011
    • Shimane University
      • Department of Radiology
      Matsue-shi, Shimane-ken, Japan
  • 2009–2011
    • University of Tehran
      • School of Electrical and Computer Engineering
      Tehrān, Ostan-e Tehran, Iran
    • University of Chicago
      • Department of Radiology
      Chicago, IL, United States
  • 1970–2011
    • Osaka University
      • • Department of Radiology
      • • Graduate School of Information Science and Technology
      • • Department of Surgery
      • • Department of Internal Medicine
      • • School of Medicine
      • • Mucosal Immunology Group
      Suika, Ōsaka, Japan
  • 2010
    • Islamic Azad University
      Teheran, Tehrān, Iran
    • Kinki University
      • Department of Radiology
      Ōsaka-shi, Osaka-fu, Japan
  • 2009–2010
    • Ritsumeikan University
      • College of Science and Engineering
      Kioto, Kyōto, Japan
  • 2005–2007
    • Sapienza University of Rome
      • Department of Radiological, Oncological and Pathological Sciences
      Roma, Latium, Italy
  • 1992–2005
    • Kyoto Institute of Technology
      • Department of Mechanical and System Engineering
      Kioto, Kyōto, Japan
  • 1988–2005
    • Osaka Police Hospital
      Ōsaka, Ōsaka, Japan
  • 2004
    • The American University of Rome
      Roma, Latium, Italy
  • 1994–2001
    • Sakurabashi Watanabe Hospital
      Ōsaka, Ōsaka, Japan
  • 1993–2001
    • Kansai Rosai Hospital
      Itan, Hyōgo, Japan
  • 2000
    • National Cerebral and Cardiovascular Center
      • Department of Cardiovascular Medicine
      Ōsaka, Ōsaka, Japan
  • 1999–2000
    • Osaka National Hospital
      Ōsaka, Ōsaka, Japan
    • University Hospitals Birmingham NHS Foundation Trust
      Birmingham, England, United Kingdom
    • University Hospital Medical Information Network
      Edo, Tōkyō, Japan
  • 1993–2000
    • Osaka Rosai Hospital
      Ōsaka, Ōsaka, Japan
  • 1998
    • Osaka Minami Medical Center
      Ōsaka, Ōsaka, Japan
  • 1984
    • Bronx-Lebanon Hospital
      Bronxville, New York, United States