IMAGE SYNTHESIS USING GENERATIVE ADVERSARIAL NETWORK TO OVERCOME IMBALANCE PROBLEMS IN CHEST X-RAY IMAGE CLASSIFICATION CASE
The imbalance of datasets in chest X-ray presents a significant challenge in building accurate and reliable pre-diagnosis models. Imbalance occurs when one label within the dataset has a much lower occurrence compared to other labels. Utilizing an imbalanced dataset for pre-diagnosis model constr...
Saved in:
| 主要作者: | |
|---|---|
| 格式: | Theses |
| 語言: | Indonesia |
| 在線閱讀: | https://digilib.itb.ac.id/gdl/view/73938 |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 機構: | Institut Teknologi Bandung |
| 語言: | Indonesia |
| 總結: | The imbalance of datasets in chest X-ray presents a significant challenge in
building accurate and reliable pre-diagnosis models. Imbalance occurs when one
label within the dataset has a much lower occurrence compared to other labels.
Utilizing an imbalanced dataset for pre-diagnosis model construction can lead to
underfitting and overfitting conditions. Although some studies have been conducted
by adapting learning algorithms, such approaches do not address the issue of
imbalanced data distribution. In this paper, we generate synthetis X-ray images
using generative adversarial network algorithms to enhance the classification
model for pneumonia infection cases. This study produces synthesis X-ray images
with lower Fréchet Inception Distance score compared to conventional data
augmentation and SMOTE. Additionally, the classification model with the addition
of synthesis data yields a significant improvement in F1 scores based on the Mann-
Whitney U test. |
|---|