Abstract
Imageless total knee arthroplasty (ITKA) eliminates the need for pre-operative imaging by reconstructing the patient-specific knee intra-operatively. Existing systems typically require a fixed number of femur and tibia patches, totalling approximately 4-7 minutes of surgeon time. In this paper we propose KneeFlow, a conditional Optimal Transport Flow Matching (OT-FM) model that generates accurate patient-specific femur and tibia point clouds using only anatomical landmarks and condylar boundary points without the need for time-consuming surface patch collection. KneeFlow integrates a Point Transformer-based boundary encoder, a landmark encoder, and a vector field network trained with an OT interpolant. By eliminating surface patch collection entirely, KneeFlow reduces intra-operative digitization from 4-7 minutes to under one minute per knee. On held-out test bones, KneeFlow achieves an Average Surface Distance (ASD) of 2.03 mm for the femur (29 test bones) and 1.29 mm for the tibia (33 test bones), outperforming GAN, Point Transformer, and Denoising Diffusion Probabilistic Model (DDPM) baselines, demonstrating the potential of proposed KneeFlow for imageless TKA. Code is available at https://github.com/IGRS-Imaging/KneeFlow
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ShapeMI_037.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=ejJ8uwspxH
BibTex
@InProceedings{SriRag_KneeFlow_MICCAISAT2026,
author = { Srinivasan, Ragavan AND Maik, Vivek AND Lakshmanan, Manojkumar AND Sivaprakasam, Mohanasankar},
title = { { KneeFlow: A Flow Matching Framework for Patient-Specific Knee Reconstruction } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17259},
month = {pending},
page = {pending}
}