| 研究課題/領域番号 |
16K00394
|
|---|
| 研究機関 | 広島大学 |
|---|
研究代表者 |
ライチェフ ビセル 広島大学, 工学研究科, 助教 (00531922)
|
|---|
| 研究期間 (年度) |
2016-04-01 – 2019-03-31
|
| キーワード | iPS cells / deep learning / segmentation / CNN / colony detection |
|---|
| 研究実績の概要 |
Induced pluripotent stem (iPS) cells have shown a huge potential to revolutionize medical therapy by personalizing regenerative medicine and creating novel disease models for research and therapeutic testing. However, since a large number of undifferentiated human iPS cells must be prepared for use as a renewable source of replacement cells for regenerative medicine, the development of an automated culture system for iPS cells is considered to be crucial. The purpose of this research project is the design of a machine learning method for automatic detection of Good/ Bad colonies of iPS cells, which would make possible to automate the cell harvesting process.
In more detail, in this project we are working on developing a new method for automatic segmentation at a pixel level of Good (undifferentiated) vs. Bad (differentiated) colonies of iPS cells directly from images of the colonies. Our method is based on deep learning using a Convolutional Neural Network (CNN) with a modified architecture which is better able to capture the relevant structural information regarding the spatial distribution of class probabilities. Experiments have shown that very promising results can be obtained by the proposed method.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
At this stage we have significantly improved our initial algorithm, developed during the previous stage, by designing several alternative cost functions, which are better suited to represent the objective to learn the relevant structure in the local patches. This has resulted in increased accuracy of the image segmentation for our task. We have conducted a series of experiments on our dataset where we compare our results to several methods which can be considered state-of-the-art for medical image segmentation and the obtained results look very promising.
|
| 今後の研究の推進方策 |
We are planning to continue working on alternative methods for representing the category-related structure in the local patches and also to consider additional representation which might be able to improve segmentation accuracy by incorporating also more globally-relevant information. We also plan to test our method on additional data sets, in order to evaluate its performance on alternative tasks (use for other more general biomedical image segmentation tasks) and also for the cases when larger training sets are available.
|
| 次年度使用額が生じた理由 |
（理由）
Presentation of our work at two international conferences had been planned for FY2017 but for reasons beyond our control we were not able to attend these.
（使用計画）
The incurrent amount will be used for presenting our work and/or information gathering at international conferences related to the current research project.
|
