Related Experiment Video
Updated: Jun 7, 2025
03:55
Computer-Aided Three-Dimensional Visualization in the Treatment of Locally Advanced Thyroid Cancer
Published on: June 9, 2023
482
Deep learning-based automatic pipeline system for predicting lateral cervical lymph node metastasis in patients with
Pengyi Yu1,2,3,4, Cai Wang1,2,3,4,5, Haicheng Zhang6
1Department of Otorhinolaryngology, Head and Neck Surgery, Yantai Yuhuangding Hospital, Qingdao University, Yantai 264000, China.
Chinese Journal of Cancer Research = Chung-Kuo Yen Cheng Yen Chiu
|November 14, 2024
Summary
A deep learning pipeline system (DLAPS) accurately diagnoses lateral lymph node metastasis in papillary thyroid carcinoma (PTC) using CT scans. This AI tool improves diagnostic accuracy and reduces unnecessary surgeries, showing strong clinical applicability.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Papillary thyroid carcinoma (PTC) diagnosis requires accurate assessment of lateral lymph node metastasis (LLNM).
- Current diagnostic methods can be invasive or lack precision, impacting treatment decisions.
Purpose of the Study:
- To develop and evaluate a deep learning-based automatic pipeline system (DLAPS) for non-invasive LLNM diagnosis in PTC using computed tomography (CT).
- To assess the DLAPS's performance against existing clinical models and radiologist performance.
Main Methods:
- A DLAPS was created using RefineNet for auto-segmentation and an ensemble model (ResNet, Xception, DenseNet) for classification.
- The system was trained and validated on 1,266 lateral lymph nodes (LLNs) from 519 PTC patients across multiple test sets.
- Performance was compared to manual segmentation DL models, clinical models, Node-RADS, and radiologist performance, with RNA-sequencing for biological insights.
Main Results:
- The DLAPS achieved high diagnostic performance with AUCs of 0.872 (internal), 0.910 (external), and 0.822 (prospective) test sets.
- It significantly outperformed clinical models (AUC 0.731) and Node-RADS (AUC 0.602).
- The DLAPS-assisted strategy improved radiologist accuracy, reduced unnecessary dissection rates from 33.33% to 7.32%, and showed associations with cell-cell conjunctions.
Conclusions:
- The DLAPS effectively segments and classifies LLNs from CT images in PTC patients non-invasively.
- The system demonstrates good generalization ability and significant clinical applicability.
- The DLAPS offers a promising tool for improving LLNM assessment in PTC, aiding clinical decision-making and reducing overtreatment.
