Trochlear dysplasia is commonly assessed using MRI, yet no standardized patient-specific preoperative plan exists to guide surgeons in reshaping the trochlear groove. In this project, we developed a generative approach based on Wavelet Diffusion Models to replace the dysplastic trochlear region in 3D knee MR images with an anatomically plausible pseudo-healthy morphology while preserving the patient’s individual anatomy, particularly the native patella. The method was trained on knee MR images from the fastMRI dataset and evaluated on 49 knee MR scans from patients with trochlear dysplasia. The generated pseudo-healthy images showed significant improvements in clinically established measures including the sulcus angle, trochlear groove depth, and Déjour classification. This work demonstrates the feasibility of generating patient-specific surgical targets directly from MRI and establishes a basis for more objective preoperative planning of trochleoplasty.