| 研究課題/領域番号 |
20K08012
|
|---|
| 研究機関 | 仙台高等専門学校 |
|---|
研究代表者 |
張 暁勇 仙台高等専門学校, 総合工学科, 准教授 (90722752)
|
|---|
| 研究分担者 |
費 仙鳳 東北文化学園大学, 科学技術学部, 准教授 (20620470)
|
|---|
| 研究期間 (年度) |
2020-04-01 – 2023-03-31
|
| キーワード | Mammograpy / Deep Learning / Interpretablity / Computer-Aided Diagnosis |
|---|
| 研究実績の概要 |
The purpose of this research is to develop an interpretable deep learning (DL)-based computer-aided diagnosis (CAD) system for breast cancer diagnosis in mammogram. According to the research plan, the research achievements in the FY2020 are summarized as follows.
(1) Tow datasets, including mammograms, radiologist annotations, and biopsy-proven diagnosis results, have been collected at Tohoku university hospital and Miyagi cancer society.
(2) Preliminary experiments have been conducted to utilize several state-of-the-art end-to-end DL models, such as faster R-CNN, SSD, and YOLO, to detect/localize the suspicious lesions in mammograms.
(3) Four papers have been published in international conferences.
|
| 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
In the FY2020, the research has been conducted and progressed smoothly according to the research plan. Two mammogram datasets obtained form Tohoku university hospital (about 300 cases) and Miyagi cancer society (about 3,000,000 cases) have been well organized for DL model usage. The data annotation is conducting with cooperation of radiologist in Tohoku university hospital. For papers related with DL-based CAD system have been published in international conferences, and two papers are planned to submit to prime international journals (scientific report, and medical physics).
|
| 今後の研究の推進方策 |
According to the research plan, the main research in FY2021 will be focused on the following three tasks. (1) Continually collecting and organizing the mammogram dataset, especially collecting the radiologist annotation, for training the DL models. (2) Due to the insufficiency and unbalance of training data, avoiding the overfitting problem and improve the training accuracy will be focused during the development of CAD system. (3) Several interpretable DL techniques, such Gradient-weighted class activation mapping (CAM), generative adversarial network (GAN), will be implemented tentatively and observe their performance for generating an interpretable diagnosis result.
|
| 次年度使用額が生じた理由 |
Due to the COVID-19 pandemic, the travel expense used for attending the domestic and international conferences was not used during the FY2020. We plan to use these part of expense for journal publication fee and international conference registration fee in the next fiscal year.
|
