2013 Fiscal Year Research-status Report
Active Learningを用いた大腸癌自動診断システム
| Project/Area Number |
25330337
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Research Institution | Hiroshima University |
|---|
Principal Investigator |
ライチェフ ビセル 広島大学, 工学(系)研究科(研究院), 助教 (00531922)
|
|---|
| Project Period (FY) |
2013-04-01 – 2016-03-31
|
| Keywords | 癌の自動診断システム / Active learning / 大腸癌 / NBI画像 / Ensemble methods |
|---|
| Research Abstract |
In developing computerized systems for automatic diagnosis of cancer from images, it is very important to have a sufficiently large data set of training images, exhibiting the whole range of variation which can be observed in the different types of cancer. However obtaining such data sets is often difficult and impractical (even though huge volumes of raw data might be easily available) due mainly to the fact that obtaining labeled samples in a proper form for use in machine learning requires the costly time and effort of busy medical experts. The aim of this research project is to investigate methods which could make it possible to drastically reduce the number of required labeled training images for cancer diagnosis, while at the same time obtaining training data sets of very good quality, by using active learning and semi-supervised learning methods. These are being developed in a concrete setting, in the context of detection of colorectal cancer from Narrow Band Imaging (NBI) images obtained through colorectal endoscopy.
|
| Current Status of Research Progress |
Current Status of Research Progress
3: Progress in research has been slightly delayed.
Reason
We have extended the original scope of this research project to include also pixel-level classification based on ensemble methods in addition to the original intention to perform patch-level classification on whole images. The expected benefits from such an extension are:
(1) the resulting method would be more general and would not depend on the concrete configuration of patterns available in the training images. The method would operate locally and therefore might be better suited for video data also;
(2) some insight is expected to be obtained regarding the question what specific local patterns actually underlie/determine the different types of cancer, which might be difficult to obtain with the whole-patch approach.
The time necessary for experimenting with suitable pixel-centered local features and also for the development of the corresponding ensemble methods-based active learning algorithms has delayed us somewhat, in the sense that still we are working on developing the different basic ingredients of the method. Once these have been developed we expect to advance at a faster pace.
|
| Strategy for Future Research Activity |
As explained in more detail above in the Progress Status section, first we need to finish the design of the pixel-centered local features which are fed into the ensemble classifiers and finalize our experiments with these features. Then we can move to the details of the design of the combination of active and semi-supervised learning in the context of the present research problem.
|
| Expenditure Plans for the Next FY Research Funding |
We were able to purchase the hardware devices necessary for the first stage of the project at somewhat lower price than expected.
The incurring amount will be used for buying hardware devices to support computationally the project at the next stage.
|
Research Products
(2 results)
