2023 Fiscal Year Final Research Report
Development of artificial-intelligence skin disease classifier and digital biopsy by using deep learning
| Project/Area Number |
21K08339
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 53050:Dermatology-related
|
|---|
| Research Institution | Ehime University (2022-2023)
University of Tsukuba (2021) |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2021-04-01 – 2024-03-31
|
| Keywords | 人工知能 |
|---|
| Outline of Final Research Achievements |
After examining the factors influencing diagnosis in both correct and incorrect examples using GradCam, it was initially thought that a decrease in accuracy would occur when not focusing on the tumor center in the heatmap. However, it was interesting to note that there was little difference in the distribution of heatmaps between correct and incorrect examples.
Subsequently, when the model was trained to focus on the center of the images, it was observed that the accuracy decreased compared to using the entire image for training. This finding aligns with the earlier results obtained using GradCam, suggesting that not only the tumor center but also the surrounding information is utilized in tumor diagnosis.
|
| Free Research Field |
皮膚科
|
| Academic Significance and Societal Importance of the Research Achievements |
GradCamの解析によるとヒートマップにおいて腫瘍中心部に注目していない場合に正答率が下がると考えていたが，興味深いことに正答例でも誤答例でもヒートマップの分布にあまり違いが見られなかった．逆の味方をすると，ヒートマップで腫瘍部分に注目していなくとも正答してしまっている画像もかなり含まれていることを示している．また，そこで画像の中央に着目するように設定して学習をさせると逆に全体を用いた場合と比べて正答率が低下することが分かった．皮膚腫瘍の判定において中央部の腫瘍部分だけでなくその周囲の情報も判定に用いられていると言うことになる．今後の機械学習におけるアノテーションの範囲にも検討が必要となる．
|
