Local Transverse-Slice-Based Level-Set Method for Segmentation of 3-D High-Frequency Ultrasonic Backscatter From...

Area of Science:

• Medical imaging
• Ultrasound technology
• Cancer diagnostics

Background:

• Accurate segmentation of lymph node parenchyma (LNP) is crucial for quantitative ultrasound (QUS) analysis in detecting metastases.
• Ultrasound attenuation and focusing effects complicate QUS analysis, necessitating methods to compensate for depth-dependent signal variations.
• Existing segmentation methods struggle with depth-related inhomogeneity in ultrasound data.

Purpose of the Study:

• To develop and validate an automatic 3-D segmentation method for lymph nodes (LNs) to improve quantitative ultrasound (QUS) analysis.
• To enable accurate detection of metastases in freshly excised human LNs.
• To compensate for ultrasound attenuation and focusing effects in LNs.

Main Methods:

• A novel, automatic, 3-D, three-phase, statistical transverse-slice-based level-set (STS-LS) method was developed, amending Lankton's approach.
• The STS-LS method utilizes gamma probability density functions to model local speckle statistics within transverse slices, mitigating depth-related inhomogeneity.
• Radio-frequency signals from perinodal fat (PNF) were restored using smoothing cubic splines prior to segmentation of LNP, PNF, and normal saline (NS).

Main Results:

• The STS-LS method achieved high accuracy on simulated data.
• On a dataset of 54 LNs from colorectal cancer patients, Dice similarity coefficients were 0.938 ± 0.025 for LNP, 0.832 ± 0.086 for PNF, and 0.968 ± 0.008 for NS, compared to manual segmentation.
• The STS-LS method demonstrated superior performance over established methods using global and local statistics.

Conclusions:

• The STS-LS method accurately handles depth-dependent effects caused by attenuation and focusing in ultrasound imaging.
• This technique enables automatic, QUS-based cancer detection in lymph nodes.
• The STS-LS method holds potential for various ultrasound imaging applications affected by depth-dependent signal variations.