1990 Fiscal Year Final Research Report Summary
A Basic Research for Automatic Examination of Stomach Tissues
| Project/Area Number |
01550308
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
電子機器工学
|
|---|
| Research Institution | Fukui University |
|---|
Principal Investigator |
TANIGUCHI Keiji Fukui University, Engineering, Professor, 工学部, 教授 (50029063)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
NAKANO Yutaka Fukui University, Engineering, Assistant, 工学部, 助手 (00111774)
|
|---|
| Project Period (FY) |
1989 – 1990
|
| Keywords | Stomach tissue image / Gland-tubule structure / Feature parameters of gland-tubules / Image processing |
|---|
| Research Abstract |
In this project, about 600 images of stomach tissues were collected by using a data acquisition system (microscope, TV camera...). Following computer-procedures (1), (2), (3), (4) were developed to analyze and recognize that kind of images.
(1) [Apparentness of gland-tubule structure] : By converting a source image into a dot pattern which indicates the positions of cell-nuclei and calculating the statistical (feature) parameters on the pattern, the apparentness of the gland-tubule structure can be evaluated. In the experiment on actual images, structured/non-structured images were distinctively evaluated.
(2) [Discrimination of the abnormal gland tubules] : Some feature parameters for discriminating between normal structures and cancered ones were proposed. These parameter values represent the irregularity of a tubular string of cells. 75-80 percent of actual images were discriminated correctly.
(3) [Measure of concentration of gland-tubules] : A matrix representation of distances between gland-tubules was devised. As the result of experiments, this representation method was efficient and useful for detecting the adjacency and the concentration of gland-tubules.
(4) [Detection of ring cells] : In order to extract nuclear regions of ring cells, one procedure based on thresholding techniques was constructed. With this procedure followed by shape analysis, most of ring cell were detectable.
The details have been reported or will be reported in relational conferences.
|
