Clinical performance of a machine learning-based model for detecting lymph node metastasis in papillary thyroid carcinoma: A multicenter study.

Abstract

Papillary thyroid carcinoma (PTC) is a common endocrine malignancy with a generally favorable prognosis, but lymph node metastasis (LNM) complicates treatment and increases recurrence risk. Current preoperative methods like neck ultrasound often miss LNM, leading to unnecessary surgeries. This study developed a non-invasive, artificial intelligence (AI)-driven predictive model for LNM using gene expression data from 157 PTC patients and validated it with qRT-PCR across 807 participants from multiple centers. The model focused on three key genes - RPS4Y1, PKHD1L1, and CRABP1 - chosen for their predictive strength. A random forest algorithm achieved high accuracy, with an AUROC of 0.992 in training and 0.911-0.953 in external validation. RPS4Y1 emerged as a standout predictor, showing the strongest distinction between metastatic and non-metastatic cases. The study also identified immune-related pathways, such as TGF-β signaling and cancer-associated fibroblast activation, as critical in metastasis. This gene expression-based model offers a non-invasive, cost-effective solution for predicting LNM, providing valuable insights to guide surgical decisions and reduce unnecessary procedures, ultimately improving patient outcomes.