3D CNN-based Deep Learning Model-based Explanatory Prognostication in Patients  with Multiple Myeloma using...

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Magnetic resonance imaging (MRI) is utilized for prognosis prediction in multiple myeloma (MM).
• Limited research has explored artificial intelligence (AI) applications for predicting MM prognosis using MRI data.

Purpose of the Study:

• To analyze whole-body diffusion-weighted MRI data using 3D convolutional neural networks (CNNs) and explainable AI (Grad-CAM).
• To predict prognosis and identify factors contributing to prediction in MM patients.

Main Methods:

• Retrospective analysis of whole-body MRI data from 142 MM patients across two centers.
• Development and validation of a 3D CNN-based deep learning model to predict 12-month progressive disease.
• External validation using a separate cohort of 31 patients.

Main Results:

• The AI model demonstrated significant stratification of progression-free survival (PFS) in both internal (2-year PFS: 91.2% vs. 61.1%) and external validation cohorts (2-year PFS: 92.9% vs. 55.6%).
• The model achieved an area under the receiver operating characteristic curve of 0.804 in external validation.
• Grad-CAM analysis indicated that MRI signals from the spleen and vertebral/pelvic bones were crucial for prognosis prediction.

Conclusions:

• Whole-body MRI analysis using 3D CNNs is an effective standalone tool for predicting MM prognosis.
• This AI-driven approach offers a novel method for prognostic assessment in MM without requiring additional clinical data.