Abstract
Quantitative analysis of cartilage thickness plays a pivotal role in the early diagnosis and monitoring of knee osteoarthritis (OA). However, conventional segmentation-based approaches often produce noisy and anatomically inconsistent thickness maps, particularly when applied to clinical-resolution MRI scans. In this paper, we propose CartiSurface, a novel implicit surface reconstruction framework that estimates cartilage thickness by learning a signed distance function (SDF) defined between the subchondral femoral and tibial bone surfaces. CartiSurface jointly predicts smooth cartilage surfaces and continuous thickness maps by enforcing geometric priors-surface spacing, parallelism, and smoothness-through a dedicated loss formulation. Our method does not rely on explicit voxel-wise cartilage labels, allowing anatomically faithful modeling even under resolution degradation. Evaluated on the OAI dataset, CartiSurface consistently outperforms state-of-the-art baselines in terms of accuracy, surface regularity, and robustness to input variability. Qualitative visualizations further highlight its ability to capture focal cartilage thinning and maintain surface continuity across the joint. These features position CartiSurface as a clinically viable tool for early OA detection, longitudinal disease monitoring, and biomechanical modeling.