Novel Preprocessing-Based Sequence for Comparative MR Cervical Lymph Node Segmentation.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate segmentation of cervical lymph nodes is crucial for diagnosing neck pathologies.
• Manual segmentation is time-consuming and requires specialized expertise.

Purpose of the Study:

• To develop and evaluate deep learning models for automatic segmentation of cervical lymph nodes in MRI.
• To enhance the speed and accuracy of pathological mass diagnosis in the neck.

Main Methods:

• Utilized a dataset of 1346 MRI slices from 64 patients.
• Employed a preprocessing model for lymph node cropping and highlighting.
• Implemented the DeepLabv3+ architecture with a ResNet-50 encoder for segmentation.
• Compared performance with and without data augmentation.

Main Results:

• Achieved high mean Intersection over Union (IoU) scores across various MRI sequences (DWI, T2, T1, T1+C, ADC).
• The DWI sequence demonstrated the highest segmentation performance.
• Augmentation generally improved IoU values across all sequences.

Conclusions:

• Deep learning models successfully segmented cervical lymph nodes with high accuracy.
• This automated approach streamlines detection and offers a promising alternative to manual segmentation in radiotherapy.
• The study is the first to use comprehensive neck MRI sequences for automated cervical lymph node segmentation.