Journal of Decision Making and Healthcare

Electronic ISSN: 3008-1572

DOI: 10.69829/jdmh

A data pre-processing with mass-preserving optimal mass transportation for brain tumor segmentation

Journal of Decision Making and Healthcare, Volume 1, Issue 1, June 2024, Pages: 1–15

TSUNG-MING HUANG

Department of Mathematics, National Taiwan Normal University, Taipei 116, Taiwan

JIA-WEI LIAO

Department of Computer Science and Information Engineering, National Taiwan University, Taipei 106, Taiwan

WEN-WEI LIN

Department of Applied Mathematics, National Yang Ming Chiao Tung University, Hsinchu 300, Taiwan

HAO-REN YAO

National Institutes of Health, Bethesda, MD, USA


Abstract

This article aims to build a framework of brain tumor segmentation for 3D MRI brain images by using the UNet-based deep learning method with optimal mass transportation (OMT) for data preprocessing. For this purpose, we develop a novel 2-phase UNet-based OMT to increase the ratio of brain tumors in the input OMT tensors. 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 computing the eigenvector of the Dice similarity matrix corresponding to the largest eigenvalue and then using it to aggregate the probability as the predicted label. Using SegResUNet with OMT input tensors to train the data for the adult glioma (Task 1), Sub-Sahara-Africa adult glioma (Task 2), and meningioma (Task 3) in the international Brain Tumor Segmentation Cluster of Challenges 2023, the Dice scores of (whole tumor, tumor core, enhanced tumor) for online validations of Tasks 1, 2, and 3 are (0.9214, 0.8823, 0.8411), (0.8747, 0.8344, 0.8267), and (0.8316, 0.8395, 0.8401), respectively. Compared with random crop pre-processing, OMT is far superior.


Cite this Article as

Tsung-Ming Huang, Jia-Wei Liao, Wen-Wei Lin and Hao-Ren Yao, A data pre-processing with mass-preserving optimal mass transportation for brain tumor segmentation, Journal of Decision Making and Healthcare, 1 (2024), 1–15