Tumor Segmentation of Ultrasonic Thyroid Image Based on Enhanced U-Net Conditional GAN
| Autor: | Yung-Lin Chan, 詹詠麟 |
|---|---|
| Rok vydání: | 2019 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 107 In order to improve the segmentation rate of thyroid tumors, our study proposes a Conditional GAN structure based on U-Net, trying to compensate for the problem of poorly labeled medical images by combining adversarial training and attention mechanisms to improve the segmentation of tumors. This experiment uses the DDTI database, and the experimental results show that the proposed method is better than the U-Net. Under the premise that the DDTI database is used, our method is better than the direct use of the FCN architecture proposed by other papers. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|