Summary Report of the SNMMI AI Task Force Radiomics Challenge 2024

Area of Science:

• Radiomics and Machine Learning in Medical Imaging
• Oncology and Hematology Research
• Artificial Intelligence in Healthcare

Background:

• A challenge was organized to compare machine learning models for predicting survival in diffuse large B-cell lymphoma (DLBCL) patients.
• The study utilized baseline 18F-FDG PET/CT radiomics datasets for prediction tasks.
• Progression-free survival (PFS) was predicted as both a binary and continuous outcome.

Discussion:

• Nineteen models predicted continuous PFS, with 6 showing performance similar to a simple linear model using SUV and TMTV.
• Twelve models predicted binary PFS, with one showing nonsignificant improvement over a simple logistic regression model.
• The results suggest limited added value of complex radiomic features and machine learning for outcome prediction in this DLBCL cohort.

Key Insights:

• Machine learning models incorporating radiomic features did not significantly outperform simpler models for PFS prediction in DLBCL.
• Performance metrics (FOMs) indicated that basic radiomic features (SUV, TMTV) were comparable to complex models.
• The study highlights the need for careful validation and comparison against baseline models in AI-driven medical research.

Outlook:

• Further research is needed to identify specific radiomic features or patient subgroups where machine learning offers significant predictive advantages.
• The findings may inform the development of more effective AI tools for personalized treatment strategies in DLBCL.
• Future challenges could focus on integrating diverse data types or exploring advanced deep learning architectures for improved prognostic accuracy.