ALGORITHMIC DETECTION OF THE BICIPITAL GROOVE IN 3D MODELS OF THE HUMERUS

Patient specific 3D printed guides for shoulder arthroplasty exist for the glenoid but currently no commercially available system exists for the humerus. A primary barrier to adoption is the process of designing a custom guide for each patient that references specific alignment landmarks. The humerus bicipital groove is often used as a reference during the implantation of a shoulder prosthesis to guide the recreation of native retroversion. [1] It has also been used as a fixation landmark for patient specific instrumentation that guides the humeral osteotomy.[2] Therefore, an automated algorithm that identifies the bicipital groove would prove useful in the implementation of humeral component 3D printed patient specific guides. Segmented DICOM images from 39 cadaveric shoulders were used to generate 3D models of the humerus. The bicipital groove was manually digitized on all 39 models. Both the canal, and trans-epicondylar axes were automatically identified using in-house software. An anatomic coordinate system was then generated using the canal (axial) axis, the trans-epicondylar (medial-lateral) axis, and their cross-product as the third (anterior-posterior) axis. Following this, 2D axial slices of the humeral head were generated, and the perimeter was then unwound into polar coordinates with the centroid located along the canal axis. A weighted average of each polar perimeter was taken, and any cysts were identified as regions where the line doubles back onto itself and removed. The second derivative of the weighted average was then calculated to permit identification of radial depressions, and the bicipital groove location was estimated using an optimized combination of the following relationships regarding the axial trace: (i) the maximum radial mean lies between the endpoints of the articular surface, (ii) 2 of the 3 largest peaks of the 2nd derivative of the radial mean are endpoints of the articular surface which contain the maximum radial mean between them, and (iii) the minimum radial mean is highly correlated to bicipital grove location. Metrics are shown in Fiigure 1. Following optimization, the line of best fit for the estimated bicipital groove location was compared to the manually digitized locations using singular value decomposition yielding the first principal component direction. In the medial-lateral and anterior-posterior directions the difference between algorithm estimate and manually digitized line of best fit midpoints was −0.38+-0.97 mm (p=0.73) and 0.02+-0.96 mm (p=0.99), respectively for all 39 humeri. On the plane formed by the two direction vectors of the lines of best fit, the angle them between was on average 2.25 & 1.81 degrees. No significant differences between the algorithm estimates and manually digitized bicipital groove locations was detected. The bicipital groove can be automatically identified with sufficient accuracy to enable its use as a reference feature for the generation of a 3D printed patient specific guide for the humerus.