A Generalized Segmentation Method in Nasopharyngeal Cancer MRI Images Based on Multidimensional Attention Mechanism and Deep Learning
Open Access
- 13 August 2024
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 579-584
Abstract
Accurate segmentation of the Nasopharyngeal carcinoma (NPC) magnetic resonance images (MRIs) is vital and essential for the effective radiotherapy and improved patient prognosis. On the one hand, manual NPC image segmentation remains the main method in clinical practices but is very time-consuming; on the other hand, great variations of the NPC lesion sizes and NPC shapes pose significant challenges on the design of the generalized segmentation of NPC MRI images. To tackle this problem, this work proposes a generalized segmentation method leveraging a multidimensional attention mechanism and deep learning techniques to enhance the accuracy and efficiency of NPC lesion segmentation. This work integrates advanced data preprocessing and a generalized segmentation network (including an optimized encoder-decoder architecture with residual connections and a multidimensional attention mechanism) to capture global and local features effectively. Simulation results verify that the proposed method improves the pixel accuracy and intersection over union (IoU) by 37.95% and 31.75%, respectively, outperforming existing techniques and offering a promising tool for clinical applications.Keywords
This publication has 8 references indexed in Scilit:
- CA-Net: Comprehensive Attention Convolutional Neural Networks for Explainable Medical Image SegmentationIEEE Transactions on Medical Imaging, 2020
- Nasopharyngeal carcinomaThe Lancet, 2019
- Deep Attentional Features for Prostate Segmentation in Ultrasound2018
- Autofocus Layer for Semantic Segmentation2018
- Squeeze-and-Excitation Networks2018
- A survey on deep learning in medical image analysisMedical Image Analysis, 2017
- Analysis of an electronic methods for nasopharyngeal carcinoma: Prevalence, diagnosis, challenges and technologiesJournal of Computational Science, 2017
- DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFsIEEE Transactions on Pattern Analysis and Machine Intelligence, 2017