Glioma grade prediction using a cross-fusion network based on unsegmented multi-sequence magnetic resonance images.

Chen Q, Wang L, Guo S, Xia H, Yang F, Zhu Y

2022 16th IEEE International Conference on Signal Processing (ICSP), 2022, pp. 447-451. doi: 10.1109/ICSP56322.2022.9965327. (Best paper award)

---

Abstract:

Accurate prediction of high- and low-grade gliomas is of great importance for making appropriate treatment plans. Although existing Radiomics and deep learning-based methods can predict glioma grade accurately with magnetic resonance (MR) images, most of them require prior segmentation of the tumors, leading to additional annotation required to build a computer-aided diagnosis system and making its application more difficult. To deal with such issue, we propose a cross-fusion network (CFNet) for classifying high- and low-grade gliomas, which fully incorporates the multiple transformation information of each single-sequence imaging data and uses the cross-fusion module to fuse multi-sequence MR image features. The fused features are fed into a classifier to predict the grade. The experimental results show that the AUC of CFNet for predicting high- and low-grade gliomas without tumor segmentation can reach 0.9769, which increases the prediction accuracy and sensitivity by 3.0% and 11.5%, respectively, compared with the state-of-the-art deep learning models. Meanwhile, the ablation experiments demonstrate that the proposed cross fusion module and multiple transformation fusion module can effectively improve the prediction performance, which is beneficial for personalized treatment of glioma.

Chen Q, Wang L, Guo S, Xia H, Yang F, Zhu Y. Glioma grade prediction using a cross-fusion network based on unsegmented multi-sequence magnetic resonance images. 
2022 16th IEEE International Conference on Signal Processing (ICSP), 2022, pp. 447-451. doi: 10.1109/ICSP56322.2022.9965327. (Best paper award)

URL: https://doi.org/10.1109/ICSP56322.2022.9965327.