An automatic femoral distance measurement algorithm on a grid environment

FEDERICO MILANO; LUCAS RITACCO; JUAN FRANCISCO GARCÍA EIJO; ADRIAN GOMEZ; MARCELO RISK; FERNÁN GONZÁLEZ BERNALDO DEL QUIRÓS; GUILLERMO MARSHALL

Choroní, Venezuela

Conferencia; Proceedings of the Second EELA-2 Conference; 2009

CIEMAT

Tumor excision with wide surgical margins is the primary treatment of aggressive or recurrent benign bone tumors and malignant bone sarcomas. This requires a surgical resection, with a potential large residual osseous defect. As diagnostic and therapeutic techniques have improved, patients with musculoskeletal sarcomas should expect increased survival rates, decreased complications and side effects, and an improved quality of life. Functional longevity of the reconstruction surgery becomes a major concern, especially in young and physically active patients. Emphasis has been placed on biologic reconstructive alternatives due to concerns involving the durability of prosthetic materials and the increasing survivorship of patients with sarcomas. Poor anatomical matching of the size between the host and the donor can significantly alter joint kinematics and load distribution, leading to articular fractures or joint degeneration. Determining the transversal size of the distal femur is critical for obtaining an appropriate allograft. Our first aim is to measure this transversal distance manually, and then automatically using a measurement technique running on grid computing technology. The second aim is to evaluate the feasibility between these methods. In this study, a total of thirty-three fresh-frozen femoral allografts were selected from the bone bank, consisting of 15 right and 18 left bones (ages from 17 to 49 years; 22 males and 11 females). Three dimensional voxelizations of all femurs were manually created from computer tomography images. This work shows an automatic technique for the measurement of the transepicondylar axis (A) distance. This measurement determines the distance, in millimeters, between the most medial point in the medial epicondyle and the most lateral point in the lateral epicondyle. These voxelizations were positioned in the same reference system and arranged in the same angle. Axial slices of each reconstruction were taken equidistantly, generating sets of nearly five hundred binary images representing each femur. More intuitively, a binary image representing a reconstruction slice can be seen as a planar point cloud. The algorithm applied to automatically obtain the A distance runs on each binary image, hence a grid parametric job is generated to process each image independently. First, the image is processed by applying a convex hull algorithm (the simple Jarvis march algorithm) to obtain the boundary of the minimal convex set containing the femur shape. In this way, the natural concave shape induced by the intercondylar fossa is avoided. Once the representation of the femur boundary is computed as a convex polygon, its maximum diameter is measured. This is accomplished by applying the "Rotating Calipers algorithm" to the convex polygon representing the femur axial view. In this way, each binary image will have associated a number representing a possible A distance. The last step of the process is finding the maximum value of all the measured distances, being this maximum the A distance we are looking for. This last step is carried out outside the grid environment, once the distance calculations for the whole set of images are finished.