HNSCC: deep learning approach bests experts in ENE diagnosis

  • Kann BH & al.
  • J Clin Oncol
  • 9 Dec 2019

  • curated by Brian Richardson, PhD
  • Univadis Clinical Summaries
Access to the full content of this site is available only to registered healthcare professionals. Access to the full content of this site is available only to registered healthcare professionals.

Takeaway

  • A deep learning algorithm is associated with superior diagnostic ability compared with expert diagnosticians for pretreatment identification of extranodal extension (ENE) in patients with head and neck squamous cell carcinoma (HNSCC).

Why this matters

  • Diagnostic imaging of ENE by clinicians is not generally reliable, necessitating new methods.

Key results

  • The deep learning algorithm had superior area under the received operative curve (AUC) compared with 2 board-certified neuroradiologists for lymph nodes from an external institution (0.84 vs 0.70 [P=.02] and 0.84 vs 0.71 [P=.01], respectively).
  • The deep learning algorithm had superior AUC for 1 board-certified neuroradiologist (0.90 vs 0.60; P<.0001 but not the other vs p=".16)," for lymph nodes from cancer genome atlas.>
  • Deep learning algorithm assistance resulted in significantly increased AUC values for lymph nodes from the Cancer Genome Atlas (P=.0003).

Study design

  • 200 lymph nodes from 144 patients were included in the study.
  • Funding: Eastern Cooperative Oncology Group-American College of Radiology Imaging Network Paul Carbone Research Fellowship Grant.

Limitations

  • Time from CT scan to surgery not standardized.
  • Most lymph nodes were ≥1 cm, limiting applicability to smaller lymph nodes.