Quantitative analysis of diffuse liver diseases by B-mode ultrasonography using neural networks
| Project/Area Number |
04670671
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
Radiation science
|
|---|
| Research Institution | Kochi Medical School |
|---|
Principal Investigator |
HISA Naofumi Kochi Medical School, Department of Radiology, Associate Professor, 放射線部, 助教授 (30129621)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
OGAWA Kouichi Hosei University, Department of Technology, Associate Professor, 工学部・電気工学科, 助教授 (00158817)
|
|---|
| Project Period (FY) |
1992 – 1993
|
| Project Status |
Completed (Fiscal Year 1993)
|
| Budget Amount *help |
¥2,000,000 (Direct Cost: ¥2,000,000)
Fiscal Year 1993: ¥600,000 (Direct Cost: ¥600,000)
Fiscal Year 1992: ¥1,400,000 (Direct Cost: ¥1,400,000)
|
|---|
| Keywords | ultrasonic diagnosis, / diffuse liver disease, / liver cirrhosis, / neural network, / computer assisted diagnosis, / quantitative evaluation / 肝臓 |
|---|
| Research Abstract |
The purpose of this reseach is to establish a system that can make quantitative evaluation of B-mode ultrasonographic images of diffuse liver diseases. The equipments used are HITACHI EUB-565A and EUB-555 ultrasound scanners and SUN workstation IPX.We chose four factors to characterize patterns of ultrasonogrphic images, the mean, the vaiance, the coefficient of variation, and the power spectrum analysis, respectively. Each data of the region of interest(32x32 matrices) were input to the workstation through Videopix. Three layred neural network(input=4 units, intermediate=4 units, output=1 unit) was constituted and the learning was performed by back propagation method.
Image data of 15 normal and 15 typical liver cirrhosis patients were used for the learning data. The output is the degree of cirrhotic change and results are figured between O(normal) and 1(abnormal). A hundred unlearned images were examined with this system. With setting threshold of 'below 0.1 as normal' and 'above 0.9 as abnormal', the accuracy was 80% for normal and 89% for abnormal. Cocerning the type of transducer(3.5MHz linear-array, 3.5MHz curved-array, and 7.5MHz linear-array) and the difference of image processing, output data were not basically interfered, though a slight difference was recognized.
The results of evaluation by the neural network showed good agreement with the clinical evaluation, and this system is believed to act as a part of computer assisted diagnosis.
|
Report
(3 results)
Research Products
(4 results)
