Automated Segmentation of Lymph Nodes on Neck CT Scans Using Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate detection of cervical lymph nodes is crucial for head and neck cancer management.
• Radiomics and AI show potential for improving lymph node diagnosis, but require automated segmentation.
• Existing methods often struggle with segmenting small lymph nodes.

Purpose of the Study:

• To develop a non-invasive deep learning (DL) algorithm for detecting and segmenting cervical lymph nodes.
• To address the challenge of segmenting small lymph nodes (5-10 mm) in CT scans.
• To create a foundational pipeline for future AI applications in nodal metastasis detection.

Main Methods:

• Utilized a dataset of 25,119 CT slices from 221 normal neck contrast-enhanced CT scans.
• Developed and evaluated multiple DL architectures, including U-Net and an adapted spatial context network.
• Focused on segmenting small cervical lymph nodes, a particularly challenging task.

Main Results:

• The developed DL algorithm achieved a Dice score of 0.8084 for lymph node segmentation.
• Demonstrated effectiveness in detecting and segmenting small cervical lymph nodes.
• The algorithm provides a robust initial step for evaluating small structures in medical imaging.

Conclusions:

• The DL algorithm successfully detects and segments small cervical lymph nodes.
• This automated segmentation framework is a key component for advanced AI in cancer diagnostics.
• The approach has potential applications for identifying early nodal metastases, even in visually normal lymph nodes.