A Collaborative Dictionary Learning Model for Nasopharyngeal Carcinoma Segmentation on Multimodalities MR Sequences.

Area of Science:

• Medical Imaging
• Oncology
• Computer Vision

Background:

• Nasopharyngeal carcinoma (NPC) is a prevalent nasopharyngeal malignancy.
• Magnetic resonance imaging (MRI) is the preferred noninvasive diagnostic tool for NPC.
• Discriminating small, infiltrative NPC from surrounding tissues in MRI is challenging.

Purpose of the Study:

• To develop a novel method for accurate localization and segmentation of NPC in MRI sequences.
• To address the challenge of differentiating NPC from adjacent normal tissues.

Main Methods:

• A voxel-wise discriminate method for NPC localization and segmentation was proposed.
• An original multiviewed collaborative dictionary classification (CODL) model was employed for refining segmentation.
• CODL reconstructs a latent space for collective multiview analysis.

Main Results:

• CODL demonstrated capability in finding a discriminative space for multiview orthogonal data on synthetic datasets.
• Experiments on real NPC data showed CODL accurately discriminated and localized NPCs of varying volumes.
• The method achieved superior NPC segmentation performance compared to benchmark techniques.

Conclusions:

• The proposed CODL method effectively segments and localizes nasopharyngeal carcinoma in MRI.
• This technique offers robust segmentation results, assisting clinicians in NPC detection.
• CODL enhances the diagnostic accuracy for nasopharyngeal carcinoma.