Temporal Evolution of Knee Osteoarthritis: A Diffusion-based Morphing Model for X-ray Medical Image Synthesis

TL;DR

This work tackles the challenge of representing KOA progression by synthesizing temporally coherent knee X-ray frames. It introduces a Diffusion-based Morphing Model (DMM) that combines a diffusion probabilistic model with a morphing/registration module and a supervision signal from a KL-2/KL-3 classifier to generate intermediate KOA frames along a geodesic path from KL-0 to KL-4. The approach yields higher PSNR and lower NMSE than baselines, and the synthesized frames substantially improve data augmentation for KOA classification, with expert radiologists confirming clinical relevance. The framework leverages the OAI dataset, demonstrates generalization capabilities, and highlights potential uses in progression visualization, early diagnosis, and CAD-assisted decision-making.

Abstract

Knee Osteoarthritis (KOA) is a common musculoskeletal disorder that significantly affects the mobility of older adults. In the medical domain, images containing temporal data are frequently utilized to study temporal dynamics and statistically monitor disease progression. While deep learning-based generative models for natural images have been widely researched, there are comparatively few methods available for synthesizing temporal knee X-rays. In this work, we introduce a novel deep-learning model designed to synthesize intermediate X-ray images between a specific patient's healthy knee and severe KOA stages. During the testing phase, based on a healthy knee X-ray, the proposed model can produce a continuous and effective sequence of KOA X-ray images with varying degrees of severity. Specifically, we introduce a Diffusion-based Morphing Model by modifying the Denoising Diffusion Probabilistic Model. Our approach integrates diffusion and morphing modules, enabling the model to capture spatial morphing details between source and target knee X-ray images and synthesize intermediate frames along a geodesic path. A hybrid loss consisting of diffusion loss, morphing loss, and supervision loss was employed. We demonstrate that our proposed approach achieves the highest temporal frame synthesis performance, effectively augmenting data for classification models and simulating the progression of KOA.

Figures (5)

• Figure 1: The structure of the classical DDPM model. The sequence from left to right shows the original X-ray image, followed by ones with increasing levels of added noise.
• Figure 2: Flowchart of the proposed approach, with black arrows representing the data flow. First, the source $x_S$ and target $x_T$ knee X-ray images are fed into the diffusion module, where the source image $x_T$ undergoes a process that adds noise to the image, incrementally increasing its noise level. Then, the source $x_S$, target $x_T$, and noised $x_t$ images are utilized to predict the noise $\hat{n}$. Following the diffusion module, the predicted noise $\hat{n}$ and the source $x_S$ image are processed through the morphing module, which applies transformations to the images via a flow field $\varphi$ to simulate structural changes. The different levels of morphing are donated by $\eta$ values. After morphing, the synthesized images $x_S{(\varphi_\eta)}$ are supervised by a supervision module to ensure they more accurately correspond to each stage of KOA. Further details are provided in Section \ref{['proposed_model']}.
• Figure 3: A standard knee plain radiograph from the database and an identified knee joint highlighted in a red box \ref{['plainXray']}. An identified knee joint \ref{['kneeROI']}.
• Figure 4: Supervision performance metrics obtained using different values of $\lambda_1$ and $\lambda_2$ for the synthesized intermediate frames $x_S(\varphi_{\eta=0.5})$ and $x_S(\varphi_{\eta=0.75})$.
• Figure 5: The box plots visualize the different performance of the selected approaches using PSNR \ref{['PSNRR']} and NMSE \ref{['NMSEE']} for the synthesized intermediate frames $x_S(\varphi_{\eta=0.25})$, $x_S(\varphi_{\eta=0.5})$, and $x_S(\varphi_{\eta=0.75})$.