Adversarial surface refinement for enhanced morphology of dental crown prediction

Automated dental crown generation using deep learning produces geometrically accurate but smooth surfaces lacking fine-scale morphological details essential for clinical acceptance. Real dental crowns require complex surface topography including functional grooves, developmental ridges, and anatomical landmarks for optimal bite function. We present an adversarial surface refinement framework that transforms smooth crown predictions into clinically acceptable surfaces with complex morphological features. Our approach employs a Surface Pattern Generative Adversarial Network that learns real occlusal surface patterns from ground truth data and applies them through personalized pattern refinement. Refinement targets the occlusal region by engraving generated patterns via constrained vertex displacement, preserving global geometry and fit. Quantitative evaluation demonstrates excellent morphological fidelity and pattern generation quality, transforming smooth predictions into surfaces with real groove patterns, cusp formations, and ridge structures. With sub-second inference times, this approach successfully bridges the gap between geometric accuracy and morphological authenticity, producing dental restorations with clinically appropriate surface complexity. The framework maintains precise crown geometry for proper fit while adding natural surface details essential for optimal dental function.