Image biomarker shows promise in CRC diagnosis

  • Skrede OJ & al.
  • Lancet
  • 1 Feb 2020

  • curated by Jim Kling
  • 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

  • An image biomarker developed through neural networks stratifies patients with stage II and III colorectal cancer (CRC) into prognostic groups that could be used to guide patient selection for adjuvant therapy.

Why this matters

  • Traditional biomarkers have not fared well in determining prognosis for early-stage CRC.

Study design

  • 10 neural networks analyzed >12,000,000 image tiles (digital scanning of conventional hematoxylin and eosin-stained tumor tissue sections) from training (n=828), tuning (n=1645), test (n=920), and validation cohorts (n=1122).
  • Good outcome patients were aged 6 years postsurgery follow-up, and no recurrence or cancer-specific death.
  • Poor outcome patients were aged
  • Patients not fitting either the good or poor category were indeterminate, and their data were used to fine-tune the model.  
  • Funding: Research Council of Norway.

Key results

  • In the validation cohort, the biomarker distinguished poor vs good prognosis (HR, 3.84; P<.0001>
  • After adjustment for known prognostic markers that were significantly predictive in univariate analyses (pN stage, pT stage, lymphatic invasion, venous vascular invasion), the biomarker remained a significant predictor (HR, 3.04; P<.0001>

Limitations

  • The biomarker has not been confirmed prospectively.