Validation of an Artificial Intelligence-Based Prediction Model Using 5 External PET/CT Datasets of Diffuse Large...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Diffuse large B-cell lymphoma (DLBCL) patient outcomes are critical for treatment stratification.
• Accurate prognostic models are essential for personalized medicine in DLBCL.
• Current prognostic tools, like the International Prognostic Index (IPI), have limitations.

Purpose of the Study:

• To validate a deep learning (DL) model for predicting 2-year time to progression in DLBCL.
• To compare the DL model's performance against the IPI and radiomic PET/CT models.
• To assess the DL model's utility in independent clinical trials.

Main Methods:

• A deep learning model was trained on maximum-intensity projections from PET/CT scans of 296 DLBCL patients.
• The model was externally validated on 836 DLBCL patients across 5 independent clinical trials.
• Performance was evaluated using Area Under the Curve (AUC) and Kaplan-Meier curves, comparing DL, IPI, and radiomic models (clinical PET and PET).

Main Results:

• The DL model achieved a significantly higher AUC (0.66) than the IPI (0.60) (P < 0.01).
• Radiomic models (clinical PET AUC 0.69, PET AUC 0.71) demonstrated superior performance to the DL model.
• The DL model consistently outperformed the IPI across all clinical trials.

Conclusions:

• The validated DL model demonstrates superior prognostic performance over the IPI for DLBCL patients.
• This AI-driven approach can predict treatment outcomes without requiring manual tumor delineation.
• While effective, the DL model's prognostic accuracy is slightly lower than advanced radiomic models.