Invited talks (Place: Tenbusu Hall)
O1-I-1(January 19, 9:30 – 9:50)
- Title :
- A Framework for Sharing Medical Image Report and Annotations using XML
- Name :
- Prof. Woei-Chyn Chu
- Affiliation :
- Department of Biomedical Engineering, National Yang-Ming University, Taiwan
- Abstract :
- Medical imaging plays an important role in clinical diagnoses. Radiologists need to make annotations on medical images to reveal abnormal findings. Digital Imaging and Communications in Medicine (DICOM) has defined a standard (DICOM Presentation state object) for handling image annotations. Both images and annotations could be presented on Picture Archiving and Communication System (PACS) viewer automatically and consistently to aid different reviewers to examine the findings. However, because many commercial PACS have not yet support full DICOM specifications, it is not an easy task to share the annotations across these imaging systems. Thus making it hard for the subsequent diagnoses. We propose a Web-based framework that aims at sharing the mammogram reports. The solution of annotation and markup display are consistent in the medical communities on the Health Promotion Administration (HPA) breast cancer screening project in Taiwan. Results and discussion will be reported in the conference.
O3-I-1(January 19, 13:30 – 13:50)
- Title :
- Ultrasound Structure Quantification for Liver Characterization
- Name :
- Prof. Po-Hsiang Tsui
- Affiliation :
- Department of Medical Imaging and Radiological Sciences, Chang Gung University, Taiwan
- Abstract :
- Acoustic structure quantification (ASQ) is a recently developed technique widely used for detecting liver fibrosis and fatty liver. Ultrasound Nakagami parametric imaging based on the Nakagami distribution has been widely used to model echo amplitude distribution for tissue characterization. We explored the feasibility of using ultrasound Nakagami imaging as a model-based ASQ technique for assessing liver fibrosis and fatty liver. At first, we investigate the relationship between changes in backscattered statistics and the concentration of fatty droplets in the liver. Fatty liver was induced in rats fed a methionine-choline-deficient diet. Results show that the ultrasound Nakagami parameter has an excellent correlation with the concentration of fatty droplets, demonstrating that ultrasound backscatter statistics depend on the degree of fatty liver in rats. Concurrently, fibrosis in rat livers was induced by an intraperitoneal injection of dimethylnitrosamine. The results showed that the Nakagami image performs well in visualizing different degrees of liver fibrosis in rats. Finally, standard ultrasound examinations were performed on human subjects with liver fibrosis. The diagnostic value of ultrasound Nakagami imaging was evaluated using receiver operating characteristic (ROC) curves. The Nakagami parameter obtained through ultrasound Nakagami imaging decreased with an increase in the METAVIR score, representing an increase in the extent of pre-Rayleigh statistics for echo amplitude distribution. The area under the ROC curve (AUROC) was 0.88 for the diagnosis of any degree of fibrosis (≧F1). Ultrasound Nakagami imaging is a model-based ASQ technique that may be beneficial for the clinical diagnosis of early liver fibrosis.
O4-I-1(January 19, 16:40 – 17:00)
- Title :
- Mathematical Foundations for 3D Reconstruction of Micro Anatomical Structures from Serial Histological Sections
- Name :
- Prof. Hidekata Hontani
- Affiliation :
- Nagoya Institute of Technology, Japan
- Abstract :
- An ongoing research project, Multidisciplinary Computational Anatomy, aims to construct computational models, which represent multiple aspects of human bodies, using sets of medical images. The aspects to be represented can be classified into four categories: (1) Spatial scale, (2) temporal dynamism, (3) vital function, and (4) pathology. For example, a model that represents a spatial scale aspect of human bodies would describe the varieties and the correlations between the structures of the micro-anatomies observed by microscope images and those of the macro-anatomies observed by MR images. One of the most important techniques for the construction of such the multidisciplinary computational anatomy models is image registration: One must explicitly make correspondences between given images, which are obtained from different modalities or which are of different patients of different ages. In this talk, we represent some topics of image/model registration. In this project, we are now constructing a model of pancreas cancer from sets of (1) in-vivo 3D MR images of KPC mice and of (2) 2D microscope images of the pancreases extracted from the mice. The in-vivo MR images of one mouse are captured several times for observing the temporal changes of tumors of the pancreas cancer. A series of the 2D microscope images are captured by slicing the processed pancreas spatially densely in order for reconstructing a 3D microscope image from the 2D images. Image registration of the 3D MR images is required for describing the temporal change of the tumors. Image registration of the 2D microscope images is required for the reconstruction of the 3D image. Image registration between the 3D MR image that was captured just before the pancreas was extracted and the 3D reconstructed microscope image is required in order for analyzing the correlations between the macro structures and micro-structures observed in each of the 3D images. In this talk, we represent these image registrations from a mathematical point of view.
O7-I-1(January 20, 10:05 – 10:25)
- Title :
- Electrical Impedance Tomography for Pulmonary Function Analysis
- Name :
- Prof. Kuo-Sheng Cheng
- Affiliation :
- Department of Biomedical Engineering / Medical Device Innovation Center, National Cheng Kung University, Taiwan
- Abstract :
- Electrical Impedance Tomography (EIT) is a novel imaging technology for industrial applications as well as medical applications. It usually applies currents to a peripheral electrode-array and then measures the resulting voltages using two-electrode or four electrode approach. From these applications and measurements, it may produce the bioimpedance images for the cross-section of body. Currently, it has found to be very useful in a variety of clinical applications, especially the pulmonary function analysis. This technique has the advantages of low-cost, real-time, and long-term monitoring. Generally, it is not easy to characterize the air distribution in dynamics and real-time for clinical diagnosis. Electrical impedance tomography may play an important role for alleviating this problem. In this paper, how to apply the electrical impedance images to quantify and analyze the distribution of air in the lungs is discussed. Some aspects of Experimental design, Measurement calibration, and 3D visualization are included. During the experiment, five layers of thoracic sections are measured under spontaneous breathing for each subject. The impedance images and flows via mouth are both synchronously acquired. Then, the images are interpolated for three dimension reconstruction and the flow signals are also interpolated for registering to the images. In the measurement calibration, the flow measurement obtained from spirometer is used as gold standard for calibration using regression method. With the correction factors, the volume in impedance image may be quantified and compared. From the experimental results, the error may be smaller than 10 % in 3D measurement, and on the contrary greater than 100% in 2D measurement. Therefore, 2D EIT is not easy to quantify the air distribution of the lung perfusion in the current clinical application. 3D EIT has great potential to provide the lung air distribution.
O7-I-2(January 20, 10:25 – 10:45)
- Title :
- Practical Interior Tomography
- Name :
- Prof. Hiroyuki Kudo
- Affiliation :
- Division of Information Engineering, Faculty of Engineering, Information and Systems, University of Tsukuba / JST-ERATO Momose Quantum-Beam Phase Imaging Project, Japan
- Abstract :
- Non-disclosure
O8-I-1(January 20, 11:05 – 11:25)
- Title :
- Fiducial-based Simultaneous Correction for Irreproducible Gantry Motion and Involuntary Patient Motion for Geometrically Uncalibratable, C-arm-based Cone-beam Computed Tomography Systems
- Name :
- Dr. Jang-Hwan Choi
- Affiliation :
- Electronics and Telecommunications Research Institute, Korea
- Abstract :
- Non-disclosure
O8-I-2(January 20, 11:25 – 11:45)
- Title :
- X-ray Imaging Techniques for Dental Applications
- Name :
- Prof. Seungryong Cho
- Affiliation :
- Department of Nuclear and Quantum Engineering, Korea Advanced Institute of Science and Technology, Korea
- Abstract :
- As flat-panel based X-ray cone-beam CT (CBCT) technology advances, its applications are fast growing in various fields. Dental applications, as catch-phrased by digital dentistry, are greatly benefitted by the CBCT technologies. Treatment planning can be more accurately constructed, patient-specific tasks can be tailored, and the existing procedures can be replaced by more robust and efficient ones. It would be desirable if all these benefits come at a low-dose cost. Additionally, CBCT images are in general prone to various image artifacts including scatter artifacts, beam-hardening artifacts, and metal artifacts. Reduction and/or correction approaches are accordingly important to secure acceptable image qualities for clinical applications. In this paper, we briefly summarize the techniques that we have developed to meet such demands in CBCT. We will also introduce a 3D manufacturing method based on CBCT images.
O9-I-1(January 20, 13:40 – 14:00)
- Title :
- Applying Transfer Method for Deep Learning from Application Viewpoint
- Name :
- Prof. Hayaru Shouno
- Affiliation :
- The University of Electro-Communications, Japan
- Abstract :
- Deep convolutional neural networks (DCNNs), which are inspired from the vision system, show good performances for the object classification task in these years, and it becomes a de facto standard feature representation method in the field of computer vision. DCNN is a kind of multi layer neural network, which can learn feature representation provided from the input of large amount of training data. Requiring the large amount of data in the medical imaging is a hard problem, so that we investigated a transfer style learning method for DCNN. The transfer method requires small amount of medical image dataset and large amount of other labeled image dataset. We introduced a simple method, that is, the initial state of the DCNN has already been trained with large amount of natural images. Thus the DCNN could be suitable for a general object recognition task. Then we train the DCNN with diffuse lung disease (DLD) pattern classification task for fine-tuning. In the result, we found the transfer-style learning shows higher performance rather than that of the DCNN trained with only DLD dataset. We also investigated the difference between the DCNN with transfer style learning and with only DLD dataset from the viewpoint of clustering in the feature space. As the result, we found the transfer style DCNN has denser cluster representation rather than that of the one trained with only DLD dataset. Thus, we conclude the pre-matured initial state is important for such novel pattern classification.
