Machine Learning Models Show Limited Value in Predicting Cancer Survival in Medical Imaging

Strategic Insights -

Researchers from Amsterdam University Medical Center have reported findings from a machine learning challenge aimed at predicting progression-free survival in patients with diffuse large B-cell lymphoma using a baseline F-FDG PET/CT radiomics dataset. While some radiomic-based machine learning models showed better performance than simple linear or logistic regression models, the differences were nonsignificant. The study suggests that the addition of sophisticated radiomic features and use of machine learning may not provide significant value in outcome prediction.

Key Takeaways:

  • The challenge aimed to predict progression-free survival (PFS) in patients with diffuse large B-cell lymphoma, either as a binary outcome (shorter than 2 years versus longer than 2 years) or as a continuous outcome (survival in months).
  • Nineteen models for predicting PFS as a continuous outcome were received from 15 teams, and external validation identified 6 models showing similar performance to that of a simple general linear reference model using SUV and total metabolic tumor volumes (TMTV) only.
  • Twelve models for predicting binary outcomes were submitted by 9 teams, and external validation showed that 1 model had higher, but nonsignificant, C-index values compared with values obtained by a simple logistic regression model using SUV and TMTV.
  • The study suggests that there may be limited or no value in the addition of sophisticated radiomic features and use of machine learning when developing models for outcome prediction.

Statistics:

  • 19 models for predicting PFS as a continuous outcome were submitted by 15 teams.
  • 12 models for predicting binary outcomes were submitted by 9 teams.
  • 6 models showed similar performance to that of a simple general linear reference model using SUV and TMTV only.
  • 1 model had higher, but nonsignificant, C-index values compared with values obtained by a simple logistic regression model using SUV and TMTV.

Sources:

  • Summary Report of the SNMMI AI Task Force Radiomics Challenge 2024
  • Journal of Nuclear Medicine, 2025
  • Soc Nuclear Medicine Inc, 1850 Samuel Morse Dr, Reston, VA 20190-5316, USA
  • Ronald Boellaard, Dept. of Radiology and Nuclear Medicine, Amsterdam University Medical Center, Cancer Center Amsterdam, Amsterdam, Netherlands