2002 Fiscal Year Final Research Report Summary
Development of the next generation CAD system for mammography
| Project/Area Number |
13555115
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 展開研究 |
|---|
| Research Field |
Measurement engineering
|
|---|
| Research Institution | Tokyo University of Agriculture and Technology |
|---|
Principal Investigator |
KOBATAKE Hidefumi Tokyo Univ. of Agri. & Tech., Graduate School of Bio-Applications and Systems Engineering, Professor, 大学院・生物システム応用科学研究科, 教授 (80013720)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
HAGIHARA Yoshihiro Univ. of Iwate, Faculty of Technology, Lecturer, 工学部, 講師 (80293009)
NAWANO Shigeru National Cancer Center Hospital East, Radiology Division, Director, 放射線部, 部長
SHIMIZU Akinobu Tokyo Univ. of Agri. & Tech., Graduate School of Bio-Applications and Systems Engineering, Associate Professor, 大学院・生物システム応用科学研究科, 助教授 (80262880)
|
|---|
| Project Period (FY) |
2001 – 2002
|
| Keywords | computer aided diagnosis / screening of breast cancer / mammography / tumor detection / cluster of microcalcifications / feature selection / 特徴量選択 |
|---|
| Research Abstract |
The performance of the computer aided diagnosis system for mammography has been greatly improved from the view points of the area under the ROC curve. The primary results are as follows.
1) Coarse calcifications can cause false microcalcification detections. Theirs detection system has been developed, which is used to eliminate false microcalcification detections caused by coarse calcifications. The performance of the revised microcalcification detection system has been greatly improved.
2) A method for discriminating between true and false microcalcifications has been developed. It uses features obtained from the Wavelet transform of original mammograms. The false positive detection rate of the new system is about one third as compared with the conventional system.
3) A large scale experiments to investigate a semi-optimal feature set for discriminating between true and false tumors have been performed. More than 490 features have been examined from the view point of the area under the ROC curve. Among them, 50 features have been selected as a semi-optimal feature set. The performance of the tumor detection system adopting the semi-optimal feature set is excellent. It has been proven to be effective in decreasing false positive detections. The false positive detection rate of the new system is as low as a half of the conventional system
|
