PathoSyn: Imaging-Pathology MRI Synthesis via Disentangled Deviation Diffusion.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Pathology

Background:

• Current generative models for medical image synthesis often struggle with feature entanglement, leading to anatomical inaccuracies.
• Existing methods using global pixel domains or binary masks can result in corrupted anatomical substrates and structural discontinuities.

Purpose of the Study:

• To introduce PathoSyn, a unified generative framework for Magnetic Resonance Imaging (MRI) image synthesis.
• To reformulate imaging-pathology as a disentangled additive deviation on a stable anatomical manifold.
• To facilitate the development of robust diagnostic algorithms and enable precision intervention planning.

Main Methods:

• PathoSyn decomposes the synthesis task into deterministic anatomical reconstruction and stochastic deviation modeling.
• A Deviation-Space Diffusion Model learns the conditional distribution of pathological residuals, capturing localized intensity variations.
• A seam-aware fusion strategy and inference-time stabilization module ensure spatial coherence and suppress boundary artifacts.

Main Results:

• PathoSyn generates high-fidelity, patient-specific synthetic datasets with preserved global structural integrity.
• The framework produces high-fidelity internal lesion heterogeneity, outperforming holistic diffusion and mask-conditioned baselines.
• Evaluations demonstrate superior perceptual realism and anatomical fidelity on tumor imaging benchmarks.

Conclusions:

• PathoSyn offers a mathematically principled pipeline for generating realistic synthetic medical images.
• The framework supports robust diagnostic algorithm development, especially in low-data regimes.
• PathoSyn enables interpretable counterfactual disease progression modeling and benchmarking of clinical decision-support systems.