A new end-to-end network model for medical image segmentation

Hongyou Chen ； Zengyong Xu

《淡江理工學刊》 24卷2期 (2021/04) Pp. 207-213

https://doi.org/10.6180/jase.202104_24(2).0009

摘要

Medical image segmentation is a necessary step to assist disease diagnosis. Due to fuzzy boundary and low contrast of human organs, medical image automatic segmentation is still a difficult problem. Aiming at the problem of poor accuracy by using the traditional fully convolutional neural network (FCN), this paper proposes a new end-to-end network model for medical image segmentation. Firstly, linear pixel value transformation is used to adjust the brightness and contrast of the original data. Then the histogram equalization is used to remove the noise and keep the main details of the image. Then the proposed FCN network is trained by using the processed data set. Finally, we make comparison with other state-of-the-art segment methods, the results show that our proposed has better segmentation effect and it can provide reliable evidence for clinical diagnosis.

關鍵字

Medical image segmentation ； FCN ； histogram equalization ； linear pixel value transformation