We recently developed an original computer aided diagnosis for dementia method that calculates the similarity of the brain disease as an objective index value and evaluated its performance in detection of Alzheimer's disease (AD) and Mild Cognitive Impairment (MCI). On the other hand, we selected about 6000 normal subjects carefully from the whole-body FDG-PET images of about 20000 examples, and made rigorous normal brain database. These two techniques were introduced into the system of Hamamatsu Medical Imaging Center and experimentally performed on a clinical setting. In comparison with the diagnosis of three doctors for 735 subjects in 2 years, the detection power of the CAD system for dementia were sensitivity 75.6%, specificity 78.1% and accuracy 77.8%, respectively. These results were consistent with our previous reports for MCI. It can be said that our method is useful as a supporting tool of diagnosis for dementia in the case of a large-scale brain medical checkup targeting normal subjects.

View full abstract

In order to develop a computer-aided diagnosis system for neonatal cerebral disorders, some methods of brain segmentation from MR images using atlas model have been studied. As neonatal cerebrum deforms quickly by natural growth, single model cannot represent growth model properly. Due to the variation of newborn brain growth even at same age, age based model will not give appropriate result. In this paper, we propose a method for estimating growth index using manifold learning and generating fuzzy object growth model (FOGM). Brain anatomical landmarks are used for manifold learning. In addition, we propose a fuzzy connectedness segmentation method using FOGM to segment the brain region. In comparison with the previous single model based method, the proposed method improved the segmentation accuracy by using FOGM.

View full abstract

In medical image analysis, the three-dimensional (3D) shape representation and modeling of anatomic structures using only a few parameters is an important issue, and can be applied to computer assisted diagnosis, surgical simulations, visualization, and many other medical applications. The conventional statistical shape models are constructed by directly analyzing the shape vector represented by a set of landmark points based on the surface information, which is also called as point distribution models (PDM). In generally PDM used vertexs of surface points for shape representation, as a shape information. However, it is difficult to find the corresponding points among the samples. Therefore, PDM cannot efficiently describe the shape information, especially the global shape information, because of its large number of vertexes. The paper presents an efficient shape representation method using spherical harmonic functions (SPHARM) for 3D anatomical structure such as the liver, spleen and so on. We show that the 3D shape of the liver can be reconstructed by a few coefficients of the SPHARM.

View full abstract