Additive manufacturing (AM), or 3D printing, is being increasingly used in medicine to manufacture tangible anatomical models or patient-specific constructs such as medical implants. Such constructs can reduce operating time and costs and enhance the accuracy of surgical procedures. The medical AM process comprises three steps:
1) image acquisition using computed tomography (CT);
2) image processing; and
3) additive manufacturing.
However, image processing remains one of the most tedious and time consuming tasks in medical AM. The most important image processing step is image segmentation: the partitioning of images into regions of interest that correspond to a specific anatomical structure (e.g., “bone”, or “tumor”). Therefore, the aim of this pathfinder project was to develop and train a convolutional neural network (CNN) for bone segmentation in CT scans.
This CNN was trained using data acquired from patients with skull defects who had previously been treated at the VUmc. The bone segmentation performance of the CNN was comparable and in certain aspects better than global thresholding (the most commonly used image segmentation method used in medical AM). The developed methodology offers the opportunity of removing the prohibitive barriers of time and effort during image segmentation, making patient-specific AM constructs more accessible.