Related Experiment Video
Updated: Apr 30, 2026
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
Deep Learning for Pretreatment Imaging-Based Tumor and Nodal Classification in Head and Neck Squamous Cell Carcinoma:
Jannik Ketschau1, Alex Grabenhorst1, Hannes Singer1
1Department of Oral and Maxillofacial Surgery, TUM University Hospital Klinikum Rechts der Isar, School of Medicine and Health, Technical University of Munich, Munich, Germany.
Background:
Accurate pretreatment assessment of the extent of tumor invasion and status of cervical lymph node metastasis is essential for staging and treatment planning in HNSCC. Deep learning (DL) shows promise but is limited by methodological heterogeneity.
Methods:
We conducted a systematic review and network meta-analysis (PRISMA). Studies (2019-2025) evaluating DL models for pretreatment lymph node or tumor invasion classification using CT, MRI, PET/CT, or SPECT/CT were included. Diagnostic performance (AUC) was pooled using random-effects models.
Results:
Twenty-three studies were included. Pooled AUC for lymph node classification was 0.78 (95% CI 0.72-0.84). In subgroup analyses, performance was lower in multicenter and externally validated studies (AUC 0.84) than in single-center studies (AUC 0.92; p = 0.029), and slightly lower with radiologic versus pathologic ground truth (0.83 vs. 0.87; p = 0.093). Network meta-analysis showed a nonsignificant advantage of fusion models. Pooled AUC for tumor invasion was 0.84. DL models outperformed human readers (ΔAUC +0.09).
Interpretation:
DL demonstrates strong diagnostic performance for pretherapeutic HNSCC imaging, but results depend on study design. Standardized datasets and prospective multicenter validation are required.