Prediction of the Repair Surface over Cartilage Defects: A Comparison of Three Methods in a Sheep Model

Manuela Kunz, Steven Devlin, Ren Hui Gong, Jiro Inoue, Stephen D. Waldman, Mark B. Hurtig, Purang Abolmaesumi, A. James StewartMICCAI 2009 · DOI

TL;DR

This paper compares three methods for predicting the original articular cartilage surface over a defect site in a sheep knee model. The goal is to improve the accuracy of osteochondral transplantation, a surgical technique for cartilage repair. A cubic spline method emerged as the most accurate and efficient, outperforming manual estimation and a statistical shape atlas.

What Problem Does This Paper Address?

Articular cartilage injuries are common and significantly impact a patient's quality of life. Osteochondral transplantation, where plugs of cartilage are moved to defect sites, is a common surgical solution. However, a critical challenge is ensuring the transplanted plugs perfectly match the original curvature of the joint surface. If the plugs are too high or too low, they can lead to increased pressure, wear, and eventual degradation of the repair. This paper addresses the problem of accurately predicting the original cartilage surface over a defect site, which is essential for precise surgical planning and optimal graft placement. This is particularly difficult because the original surface is absent at the defect.

Methods

The researchers evaluated three distinct methods for predicting the missing cartilage surface:

  • Manual Estimation: An experienced surgeon used custom software to manually place virtual cartilage plugs to reconstruct the articular surface. The surgeon's goal was to best match the expected pre-defect surface.
  • Cubic Spline Surface: This method employed a bicubic Hermite surface patch. Four points on the post-defect cartilage surface surrounding the defect were selected, and the surface was mathematically defined using these points and their derivatives. This method is parametric and aims to create a smooth, continuous surface.
  • Statistical Shape Atlas: This approach involved building a statistical model of normal distal femur shapes from 12 sheep cadaver knees. This "atlas" was then fitted to the post-defect CT scan of the target knee using an optimization algorithm (CMA-ES) to predict the missing cartilage surface.

All three methods were tested on 16 sheep knees that had surgically induced cartilage defects. The accuracy of each predicted surface was then compared to the "gold standard" – the original cartilage surface from a pre-defect CT scan of the same knee.

Datasets

  • Training Data: 12 sheep cadaver femurs were used to construct the statistical shape atlas. These femurs were scanned in air and manually segmented to create binary CT volumes.
  • Evaluation Data: 16 sheep knees were used for the in-vivo study. Each knee underwent an initial arthro-CT scan to capture the pre-defect cartilage surface. A cartilage defect was then induced, and three months later, a second arthro-CT scan was performed to capture the post-defect state.

Key Insights and Ideas

  • The importance of surface congruity: The study strongly reinforces that precise matching of the repaired cartilage surface to the original joint geometry is paramount for successful long-term outcomes in osteochondral transplantation.
  • Splines offer a sweet spot: The cubic spline method strikes a good balance between accuracy, consistency, and computational efficiency, making it a promising candidate for automated surgical planning.
  • Atlas limitations: The statistical shape atlas, while conceptually powerful, showed poorer performance in this specific application. This highlights that the quality and size of the training data, as well as the fitting process, are critical for atlas-based methods.
  • Manual estimation is time-consuming: While performed by an expert, manual placement of plugs is labor-intensive and subjective, suggesting a need for more automated solutions.

Key Results

  • Accuracy: The cubic spline surface was found to be significantly more accurate than the statistical shape atlas (p < .0001). The manual surface was also more accurate than the atlas (p < .0001). The manual and spline surfaces had equivalent accuracy, with an effect size of 0.25 mm RMS error considered insignificant.
  • Consistency: The spline surface exhibited a smaller variance in RMS error and a much smaller maximum RMS error compared to the manual surface.
  • Efficiency: Operators estimated the following times to construct the predicted surface: 5 minutes for the spline, 20 minutes for manual estimation, and 110 minutes for the atlas (including optimization iterations).

Limitations and Future Work

  • Atlas Performance: The authors suggest that the poor performance of the atlas might be due to the construction or fitting process. They propose that a larger training dataset (more than 12 knees) and restricting the atlas to only the affected condyle could improve matching.
  • Manual Estimation Time: While accurate, the manual method is time-consuming, and the time estimate for the atlas included significant optimization iterations.
  • Future Work: The paper suggests that a spline surface has significant advantages for automated planning systems due to its simplicity and speed. Further development of automated planners for mosaic arthroplasty is envisioned, with a spline surface appearing to be a strong candidate for representing the predicted cartilage repair surface.

Summary updated

BibTeX
@inproceedings{kunz2009prediction,
  author = {Manuela Kunz and Steven Devlin and Ren Hui Gong and Jiro Inoue and Stephen D. Waldman and Mark B. Hurtig and Purang Abolmaesumi and A. James Stewart},
  title = {{Prediction of the Repair Surface over Cartilage Defects: A Comparison of Three Methods in a Sheep Model}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2009},
  doi = {10.1007/978-3-642-04268-3\_10},
}
Figure