Retinal fundus images are useful for the early detection of a number of ocular diseases. Examinations using retinal fundus images are cost effective and are suitable for mass screening. Consequently, retinal fundus images are obtained in many health care centers and medical facilities during medical checkups for ophthalmic examinations. The increase in the number of ophthalmic examinations improves ocular health care in the population; however, it also increases the workload of ophthalmologists. Therefore, computer-aided diagnosis (CAD) systems developed for analyzing retinal fundus images can assist ophthalmologists by reducing their workload and in improving the screening accuracy. This paper describes the background of CAD aimed at detection and analysis of ocular diseases such as diabetic retinopathy, hypertensive retinopathy, and glaucoma using retinal fundus images. Moreover, this paper reviews the CAD systems reported from domestic and overseas research groups, including our group. We believe that the CAD systems have the potential to help ophthalmologists in their diagnosis.

View full abstract

The recent development of medical, radiological and hospital information systems has brought evolution in radiology. However, the development regarding usability of the medical information system has not been enough carried out. The system sometimes disturbs smooth operations in medical examination. Therefore, a new expert system was applied to RIS (Radiology Information System) , in which the usability was improved for medical staffs. The expert system named KIOS (Knowledge Interactive Organizing System) is based on the Case-Based Reasoning. This expert system has two main features, that is, a knowledge model and a reasoning engine. The knowledge model is a network type one having variable weight to each network. The reasoning engine has adaptively organizing function which automatically controls the weight of network in the knowledge model and changes the knowledge network to a new one during interactive operations of KIOS. We developed an expert server and RIS client simulator using C # language and Microsoft .net framework. As the result, this expert system was executed successfully that the average response time was 0.283 sec for one client connection and 0.47 sec for simultaneous five client connections. In the study, it was found that the KIOS expert system is very useful for improving the efficiency in practical use of RIS.

View full abstract

We have reported the usefulness of a fully automated dynamic subtraction method of hepatic CT by use of 3-D phase only correlation method for improvement of detection of hepatic tumors. However, when the liver moves greatly in breathing, subtraction of CT images fails due to mis-registration of livers in this method. So, we have developed an improved fully automated dynamic subtraction method of hepatic CT. In this method, we used segmented extracted liver regions instead of extracted bone regions. For extracting liver regions from CT images, we used a level set method. The performance of the improved method was superior to that of our previous method, especially, in the cases those livers moved greatly in breathing.

View full abstract

The dependence of CT number and image contrast on the tube voltage, namely the photon energy in diagnostic CT imaging was theoretically investigated. The linear attenuation coefficients for various elements and mixtures were obtained for three cases that 1) for the bremsstrahlung X-rays with an ideal detector, 2) for the bremsstrahlung X-rays with an assumed detector, and 3) for the monoenergetic X-rays, by using the X-ray photon spectrum and the sensitivity spectrum of the CT detector. As a result, the CT number to several materials converged to zero with an increase of tube voltage in all cases, whereas that of some materials diverged from zero. This result indicates that the image contrast of some materials in the restricted energy range such as diagnostic range may be emphasized by using high tube voltage, as opposed to common knowledge.

View full abstract

Bone scintigraphy is the most effectively applied examination in finding temporal transfer of cancerous diseases such as breast cancer to bone. Although diagnosis of bone scintigrams has been based commonly on inspection by eyes, it may be more informative by introducing quantitative evaluation and automatic detection of abnormal concentration of RI agents. For this purpose, a first task has been set as noise removal, because the high noise level of the scintigrams can be obstructive to machine processing. Under the condition of low resolution, we have studied the feasibility of anisotropic diffusion and its variant of the total variation (TV) minimization method. Several methods have been applied to noisy test patterns in order to determine optimal values of the parameters. The Tsuji diffusion with and without regularization and the TV minimization resulted promising with a few thousands of repetitions with minimal errors and fair visual appearances. When applied to clinical bone scintigrams, they resulted promising visual appearances with about a hundred repetitions.

View full abstract