TCIA H&N1 HNSCC (MAASTRO)

Overview

The H&N1 collection comprises 136 head and neck squamous cell carcinoma (HNSC) patients treated at MAASTRO Clinic (Maastricht, Netherlands). It served as the first HNSCC validation cohort in Aerts et al. 2014, testing cross-disease transferability of the NSCLC-trained radiomic prognostic signature. CT scans and survival data are hosted on TCIA. PMID:24892406

Composition

  • Cancer type: HNSC (head and neck squamous cell carcinoma), n=136.
  • Modality: CT; head-and-neck immobilisation protocol (reduced motion artefact vs. free-breathing lung CT).
  • Annotations: manual tumour delineations; HPV status available for a subset.
  • Clinical data: treatment records, TNM staging, overall survival.
  • Role in Aerts 2014: first cross-disease validation of the NSCLC-derived four-feature radiomic signature. PMID:24892406

Assays / panels (linked)

  • CT radiomics — 440-feature library applied with Lung1-derived feature definitions.

Papers using this cohort

  • PMID:24892406 — Aerts et al. 2014, Nature Communications: H&N1 used as first HNSCC validation; the four-feature signature achieved concordance index 0.69 (P=7.99 × 10⁻⁷, Wilcoxon).
  • PMID:37397861 — Kim et al. 2023, Radiotherapy and Oncology: HN1 used as external validation cohort (n=137 oropharynx + larynx); validated in-house at MAASTRO using pretrained challenge models.

Notable findings derived from this cohort

  • 135 / 440 radiomic features (31%) were significantly prognostic in H&N1 at FDR 10% using Lung1-derived median thresholds. PMID:24892406
  • Signature + TNM significantly improved over TNM alone; signature performance was comparable to TNM staging in H&N1. PMID:24892406
  • No significant association between the radiomic signature and HPV status (P=0.17, combined H&N1+H&N2), supporting the signature’s complementarity to HPV-based stratification. PMID:24892406
  • In the HNSC prognostic challenge, the overall model ranking was fairly consistent with internal test results on HN1; engineered radiomics + EMR (Model 3) and deep radiomics + EMR (Model 5) performed better on HN1 than on RADCURE but still did not beat the top MTLR + EMR model. PMID:37397861

Sources

  • TCIA collection: Head-Neck-PET-CT — https://www.cancerimagingarchive.net/collection/head-neck-pet-ct/
  • PMID:24892406 — Aerts et al. 2014, Nature Communications, DOI 10.1038/ncomms5006.
  • PMID:37397861 — Kim et al. 2023, Radiotherapy and Oncology.

This page was processed by crosslinker on 2026-05-04.