Explainable AI-Assisted Multimodal Ultrasound Radiomics for Preoperative Risk Stratification of Central Lymph Node...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Papillary thyroid carcinoma (PTC) frequently involves central lymph node metastasis (CLNM), impacting treatment decisions.
• Accurate preoperative risk stratification of CLNM is crucial for effective patient management.

Purpose of the Study:

• To develop and validate an explainable AI-based multimodal approach for preoperative CLNM risk stratification in PTC.
• To assess the impact of this explainable AI on radiologist decision-making.

Main Methods:

• A multicenter retrospective study integrated preoperative ultrasound, elastography, and clinical data.
• SHapley Additive exPlanations (SHAP) provided model interpretability.
• A reader study evaluated radiologist performance with and without AI assistance.

Main Results:

• The multimodal AI model achieved high performance (AUCs 0.975, 0.917, 0.844 in training, validation, external test sets).
• Key predictors included age, radiomic, and elastography features.
• Explainable AI significantly improved diagnostic accuracy, confidence, and interreader agreement.

Conclusions:

• An explainable AI-based multimodal approach offers accurate preoperative risk stratification for CLNM in PTC.
• This AI tool enhances radiologist performance and supports clinical decision-making in radiology workflows.