Prediction of soft tissue sarcoma response to radiotherapy using longitudinal diffusion MRI and a deep neural network...

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Predicting soft tissue sarcoma response to radiotherapy is crucial for treatment optimization.
• Longitudinal diffusion-weighted MRI (DWI) offers potential for non-invasive monitoring of treatment effects.

Purpose of the Study:

• To predict soft tissue sarcoma response to radiotherapy (RT) using longitudinal DWI.
• To investigate a deep learning framework with generative adversarial network (GAN)-based data augmentation for response prediction.

Main Methods:

• Thirty soft tissue sarcoma patients underwent longitudinal DWI during RT.
• Pathologic treatment effect (TE) scores were used to classify response.
• A VGG-19 based deep learning model with ACGAN data augmentation was trained and validated.

Main Results:

• The deep learning model achieved high training (86.5%) and validation (84.8%) accuracies.
• Slice-based prediction accuracy ranged from 81.6% to 86.8% (overall 83.3%).
• Patient-based prediction achieved a mean accuracy of 97.1%.

Conclusions:

• Deep learning models can effectively predict pathologic treatment effect from longitudinal DWI in soft tissue sarcomas.
• The developed framework shows promise for personalized radiotherapy treatment planning.