Survival prediction in diffuse large B-cell lymphoma patients: multimodal PET/CT deep features radiomic model...

Area of Science:

• Radiomics and Medical Imaging
• Oncology
• Artificial Intelligence in Medicine

Background:

• Diffuse large B-cell lymphoma (DLBCL) is an aggressive non-Hodgkin lymphoma.
• Accurate prognostic models are crucial for effective patient management and treatment strategies.

Purpose of the Study:

• To develop a predictive model for diffuse large B-cell lymphoma (DLBCL) patient survival.
• To utilize a deep features radiomics signature (DFR-signature) derived from multimodal PET-CT images.

Main Methods:

• A deep learning fusion network was used to create multimodal PET-CT images from 369 DLBCL patients.
• Deep features were extracted and a DFR-signature was built using Automated machine learning (AutoML).
• A combined model integrated the DFR-signature with clinical factors for survival prediction (PFS and OS).

Main Results:

• A DFR-signature comprising 1000 deep features was developed.
• The combined model demonstrated strong performance in predicting progression-free survival (PFS) and overall survival (OS).
• High concordance index (C-index) values were achieved in both training and validation cohorts for PFS and OS.

Conclusions:

• The DFR-signature derived from multimodal PET-CT images enhances prognostic classification accuracy in DLBCL.
• The combined DFR-signature and NCCN-IPI show significant potential for DLBCL patient risk stratification.