Abstract
The use of 3D imaging methodologies in orthopaedics has allowed the introduction of new technologies, such as the design of patient-specific implants or surgical instrumentation. This has introduced the need for high accuracy, in addition to a correct diagnosis. Until recently, little was known about the accuracy of MR imaging to reconstruct 3D models of the skeletal anatomy. This study was conducted to quantify the accuracy of MRI-based segmentation of the knee joint.
Nine knees of unfixed human cadavers were used to compare the accuracy of MR imaging to an optical scan. MR images of the specimens were obtained with a 1.5T clinical MRI scanner (GE Signa HDxt), using a slice thickness of 2 mm and a pixel size of 0.39 mm × 0.39 mm. Manual segmentation of the images was done using Mimics® (Materialise NV, Leuven, Belgium). The specimens were cleaned using an acetone treatment to remove soft-tissue but to keep the cartilage intact. The cleaned bones were optically scanned using a white-light optical scanner (ATOS II by GOM mbH, Braunschweig, Germany) having a resolution of 1.2 million pixels per measuring volume, yielding an accuracy of 0.02 mm. The optical scan of each bone reflects the actual dimensions of the bone and is considered as a ground truth measurement. First, a registration of the optical scan and the MRI-based 3D reconstruction was performed. Then, the optical scan was compared to the 3D model of the bone by calculating the distance of the vertices of the optical scan to the reconstructed 3D object.
Comparison of the 3D reconstruction using MRI images and the optical scans resulted in an average absolute error of 0.67 mm (± 0.52 mm standard deviation) for segmentation of the cartilage surface, with an RMS value of circa twice the pixel size. Segmenting the bone surface resulted in an average absolute error of 0.42 mm (± 0.38 mm standard deviation) and an RMS error of 1.5 times the pixel size. This accuracy is higher than reported previously by White, who compared MRI and CT imaging by looking at the positioning of landmarks on 3D printed models of the segmented images using a calliper [White, 2008]. They reported an average accuracy of 2.15 mm (± 2.44 mm) on bone using MRI images. In comparison, Rathnayaka compared both CT- and MRI-based 3D models to measurements of the real bone using a mechanical contact scanner [Rathnayaka, 2012]. They listed an accuracy of 0.23 mm for MRI segmentation using five ovine limbs.
This study is one of the first to report on the segmentation accuracy of MRI technology on knee cartilage, using human specimens and a clinical scanning protocol. The results found for both bone and cartilage segmentation demonstrate the feasibility of accurate 3D reconstructions of the knee using MRI technology.
