Ultrasound-Based Deep Learning Radiomics to Predict Cervical Lymph Node Metastasis in Major Salivary Gland Carcinomas

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Cervical lymph node metastasis (CLNM) significantly impacts prognosis and treatment for major salivary gland carcinomas (MSGCs).
• Accurate noninvasive prediction of CLNM is crucial for surgical planning and patient management.
• Current noninvasive prediction methods for CLNM in MSGCs have limitations.

Purpose of the Study:

• To develop and validate a deep learning (DL) radiomics model using ultrasound (US) for noninvasive prediction of CLNM in MSGCs.
• To assess the model's performance in identifying CLNM preoperatively.
• To support personalized surgical approaches and reduce interventions based on accurate metastasis prediction.

Main Methods:

• A cohort of 214 patients with MSGCs from four medical centers was used, divided into training and validation sets.
• Radiomics and DL features were extracted from preoperative US images.
• A logistic regression model integrating clinical, US, radiomics, and DL features was developed and validated.

Main Results:

• The developed model demonstrated robust performance in predicting CLNM, achieving an area under the receiver operating characteristic curve (AUC) of 0.962 in the validation cohort.
• Key features including patient age, tumor edge, calcification, and US-reported CLN status were identified.
• The final model showed high accuracy (0.886), precision (0.762), recall (0.842), and F1 score (0.8).

Conclusions:

• An ultrasound-based deep learning radiomics model accurately predicts preoperative CLNM in MSGCs.
• This noninvasive approach can guide personalized treatment strategies for MSGC patients.
• The model aids in optimizing surgical decisions and potentially avoiding unnecessary interventions.