Related Experiment Video
Updated: Jul 26, 2025
08:18
Analysis of Lymph Node Volume by Ultra-High-Frequency Ultrasound Imaging in the Braf/Pten Genetically Engineered Mouse Model of Melanoma
Published on: September 8, 2021
3.0K
Universal detection and segmentation of lymph nodes in multi-parametric MRI
Tejas Sudharshan Mathai1, Sungwon Lee2, Thomas C Shen2
1National Institutes of Health (NIH) Clinical Center, Bethesda, MD, USA. tejas.mathai@nih.gov.
Summary
A new computer-aided pipeline accurately detects and segments lymph nodes in multi-parametric MRI scans. This method improves upon existing approaches by effectively utilizing T2 fat-suppressed and diffusion-weighted imaging series for better lymphadenopathy assessment.
Area of Science:
- Medical Imaging
- Radiology
- Artificial Intelligence in Medicine
Background:
- Accurate lymph node (LN) measurement in multi-parametric MRI (mpMRI) is crucial for staging metastatic disease.
- Existing methods for LN detection and segmentation in mpMRI have limitations in performance and do not fully leverage complementary imaging sequences.
Purpose of the Study:
- To develop a computer-aided detection and segmentation pipeline for universal LN identification in mpMRI.
- To improve the accuracy and reliability of LN assessment in mpMRI studies.
Main Methods:
- A pipeline was developed leveraging T2 fat-suppressed (T2FS) and diffusion-weighted imaging (DWI) series from mpMRI.
- Co-registration and blending of T2FS and DWI series were performed using selective data augmentation.
- A Mask RCNN model was trained for 3D LN detection and segmentation.
Main Results:
- The proposed pipeline achieved high precision ([Formula: see text]%) and sensitivity ([Formula: see text]%) at 4 false positives per volume.
- A dice score of [Formula: see text]% demonstrated effective segmentation performance.
- Significant improvements were observed compared to current approaches, with [Formula: see text]% increase in precision, [Formula: see text]% in sensitivity, and [Formula: see text]% in dice score.
Conclusions:
- The developed pipeline successfully detected and segmented both metastatic and non-metastatic lymph nodes in mpMRI.
- The model demonstrated flexibility, accepting either T2FS series alone or a blend of T2FS and DWI series.
- This approach removes the dependency on having both T2FS and DWI series available, overcoming limitations of prior work.
