MPM-DCE: Multi-stage Progressive Mechanism for Early and Late DCE Prostate MRI Synthesis.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Dynamic Contrast-Enhanced MRI (DCE-MRI) is vital for prostate cancer diagnosis.
• Current deep learning methods for DCE-MRI synthesis use single-stage architectures, limiting feature refinement.
• Preserving perfusion details is essential for accurate diagnostic information.

Purpose of the Study:

• To develop a multi-stage Generative Adversarial Network (GAN)-based framework (MPM-DCE) for synthesizing early and late response DCE-MRI images.
• To improve the progressive refinement of perfusion features in DCE-MRI synthesis.
• To enhance prostate cancer diagnosis by reducing reliance on contrast agents.

Main Methods:

• Proposed MPM-DCE, a multi-stage GAN framework with an encoder-decoder generator.
• Utilized Adaptive Inter-Cascade Feature fusion (AICF) blocks for sequential and lateral information exchange between stages.
• Employed multimodal inputs for synthesizing DCE-MRI images.

Main Results:

• MPM-DCE achieved state-of-the-art performance on the ProstateX dataset.
• Demonstrated superior results compared to DCE-Former, with significant improvements in PSNR, SSIM, and MAE for both early and late responses.
• Validated the importance of multi-stage progressive feature refinement and inter-stage attention.

Conclusions:

• MPM-DCE significantly improves DCE-MRI synthesis quality, preserving crucial perfusion details.
• The method reduces the need for gadolinium contrast agents, potentially enhancing prostate cancer diagnostics.
• Statistically significant improvements suggest clinical relevance for improved diagnostic accuracy and reduced unnecessary biopsies.