A convolutional neural network combined with positional and textural attention for the fully automatic delineation of primary nasopharyngeal carcinoma on non-contrast-enhanced MRI

Lun M Wong; Qi Yong H Ai; Darren M C Poon; Macy Tong; Brigette B Y Ma; Edwin P Hui; Lin Shi; Ann D King

doi:10.21037/qims-21-196

A convolutional neural network combined with positional and textural attention for the fully automatic delineation of primary nasopharyngeal carcinoma on non-contrast-enhanced MRI

Quant Imaging Med Surg. 2021 Sep;11(9):3932-3944. doi: 10.21037/qims-21-196.

Authors

Lun M Wong¹, Qi Yong H Ai¹, Darren M C Poon², Macy Tong², Brigette B Y Ma², Edwin P Hui², Lin Shi¹, Ann D King¹

Affiliations

¹ Department of Imaging and Interventional Radiology, The Chinese University of Hong Kong, Prince of Wales Hospital, Hong Kong, China.
² Department of Clinical Oncology, State Key Laboratory of Translational Oncology, The Chinese University of Hong Kong, Prince of Wales Hospital, Hong Kong, China.

Abstract

Background: Convolutional neural networks (CNNs) have the potential to automatically delineate primary nasopharyngeal carcinoma (NPC) on magnetic resonance imaging (MRI), but currently, the literature lacks a module to introduce valuable pre-computed features into a CNN. In addition, most CNNs for primary NPC delineation have focused on contrast-enhanced MRI. To enable the use of CNNs in clinical applications where it would be desirable to avoid contrast agents, such as cancer screening or intra-treatment monitoring, we aim to develop a CNN algorithm with a positional-textural fully-connected attention (FCA) module that can automatically delineate primary NPCs on contrast-free MRI.

Methods: This retrospective study was performed in 404 patients with NPC who had undergone staging MRI. A proposed CNN algorithm incorporated with our positional-textural FCA module (A_proposed ) was trained on manually delineated tumours (M_1st ) to automatically delineate primary NPCs on non-contrast-enhanced T2-weighted fat-suppressed (NE-T2W-FS) images. The performance of A_proposed , three well-established CNNs, Unet (A_unet ), Attention-Unet (A_att ) and Dense-Unet (A_dense ), and a second manual delineation repeated to evaluate human variability (M ₂ _nd ) were measured by comparing to the reference standard M ₁ _st to obtain the Dice similarity coefficient (DSC) and average surface distance (ASD). The Wilcoxon rank test was used to compare the performance of A_proposed against A_unet , A_att , A_dense and M ₂ _nd .

Results: A_proposed showed a median DSC of 0.79 (0.10) and ASD of 0.66 (0.84) mm. It performed better than the well-established networks A_unet [DSC =0.75 (0.12) and ASD =1.22 (1.73) mm], A_att [DSC =0.75 (0.10) and ASD =0.96 (1.16) mm] and A_dense [DSC =0.71 (0.14) and ASD =1.67 (1.92) mm] (all P<0.01), but slightly worse when compared to M ₂ _nd [DSC =0.81 (0.07) and ASD =0.56 (0.80) mm] (P<0.001).

Conclusions: The proposed CNN algorithm has potential to accurately delineate primary NPCs on non-contrast-enhanced MRI.

Keywords: Texture; convolutional neural network (CNN); head and neck; magnetic resonance imaging (MRI); nasopharyngeal carcinomas (NPCs).