講 題：An UNet-based Brain Tumor Segmentation Framework via Optimal Mass Transportation Pre-processing
講 師： 師大數學系 黃聰明 教授
時 間： 2023年 10月 26 日 (星期四) 15:10-17:00
In this talk, we aim to build a framework for MRI images of brain tumor segmentation using the deep learning method. For this purpose, we develop a novel 2-Phase UNet-based OMT framework to increase the ratio of brain tumors using optimal mass transportation (OMT). Moreover, due to the scarcity of training data, we change the density function by different parameters to increase the data diversity. For the post-processing, we propose an adaptive ensemble procedure by solving the eigenvectors of the Dice similarity matrix and choosing the result with the highest aggregation probability as the predicted label. The Dice scores of the whole tumor (WT), tumor core (TC), and enhanced tumor (ET) regions for online validation computed by SegResUNet were 0.9214, 0.8823, and 0.8411, respectively. Compared with random crop pre-processing, OMT is far superior.