Random Forest-Based Detection of Metastases in Clinically Scanned Lymph Nodes Using Quantitative Ultrasound Imaging

Area of Science:

• Medical imaging
• Ultrasound technology
• Oncology diagnostics

Background:

• Quantitative ultrasound (QUS) imaging analyzes radiofrequency echo signals to assess tissue microstructural properties.
• Previous ex vivo studies utilized QUS for detecting lymph node (LN) metastases.
• A need exists for in vivo methods to detect LN metastases using clinical ultrasound equipment.

Purpose of the Study:

• To develop and evaluate a quantitative ultrasound (QUS)-based method for detecting lymph node (LN) metastases in vivo.
• To assess the feasibility of using a clinical ultrasound scanner for this application.

Main Methods:

• Collected parallel radiofrequency (RF) frames from 46 cervical and axillary lymph nodes (LNs) in 45 patients.
• Computed QUS parameters (backscatter coefficient and envelope statistics) and radiomics features.
• Trained random forest models using QUS and radiomics features to classify metastatic LNs.

Main Results:

• Significant differences in average QUS parameters and radiomics features were observed between metastatic and benign LNs (p≤10⁻⁴).
• The best random forest model achieved an area under the receiver-operating characteristic curve of 0.91.
• The model demonstrated 100% sensitivity and 67% specificity in identifying metastatic LNs.

Conclusions:

• Quantitative ultrasound (QUS) imaging with a clinical scanner shows promise for in vivo detection of metastatic lymph nodes (LNs).
• This technique could aid clinicians in performing more selective lymph node biopsies or excisions.
• Further development could enhance diagnostic accuracy in cancer staging.