Automatic data-driven design and 3D printing of custom ocular prostheses

Millions of individuals require custom ocular prostheses due to eye loss or congenital defects. Traditional manufacturing, performed by highly skilled ocularists, is time-consuming and yields inconsistent quality. While additive manufacturing (3D printing) offers a promising alternative, existing approaches primarily automate the digital design process, still requiring significant expertise and time. This research addresses the need for a fully automated, data-driven process that streamlines the creation of custom ocular prostheses, improving efficiency and reproducibility while maintaining high cosmetic quality. The loss of an eye impacts both vision and appearance, stemming from trauma, painful blindness, or untreatable eye tumors. Following enucleation or evisceration, an orbital implant is inserted, and a prosthesis is worn over the conjunctiva. While pre-manufactured stock eyes exist, custom prostheses are crucial for optimal fit and appearance matching. Approximately 0.1% of the global population uses prosthetic eyes, highlighting the significant need for improved manufacturing techniques that prioritize cosmesis, comfort, and timely production.

Traditional methods for creating ocular prostheses include glassblowing and the use of polymethyl methacrylate (PMMA). Glassblowing can produce lifelike results but is delicate and requires frequent replacement. PMMA prostheses are more durable, but their appearance may be less natural. In recent years, additive manufacturing has emerged as a potential solution. Existing 3D printing approaches for ocular prostheses vary in their design process. Some rely on manual CAD design based on CT scans or 3D scans of existing prostheses or molds, while others use average prosthesis shapes or manually segmented data. Automated shape design processes have been proposed, but these often involve computed tomography (CT) scans, which present radiation risks. Crucially, most existing methods lack automated appearance recreation, relying on manual painting or limited automation of image processing. These limitations often result in small sample sizes and lack comprehensive clinical testing.

This study details a fully automated, data-driven process. The process begins by scanning the patient's fellow eye and eye socket using a modified AS-OCT device equipped with a color camera. A conformer, one of 12 pre-printed shapes representing common socket forms, is inserted to help maintain eyelid position and serve as a reference for alignment during scanning. The AS-OCT captures a volumetric image stack of the socket and a color image of the fellow eye. The ocularist selects two veining parameters (thickness and branching ratio) based on the patient’s fellow eye. The data is then processed by custom software. A statistical shape model (SSM), trained on 173 manually crafted prostheses, is used to predict the optimal prosthesis shape based on the visible portion of the socket captured by the AS-OCT. The SSM minimizes an energy function that balances the fit to the visible socket surface and the similarity to a base shape (selected conformer). The software synthesizes appearance information from the color image of the fellow eye. This involves denoising, color characterization, specular highlight removal, iris and pupil boundary detection, and sclera segmentation. A procedural method generates scleral veining based on the ocularist-selected parameters. The iris and sclera textures, along with the predicted shape, are combined to create a textured 3D model. This model is then printed using a multi-material full-color 3D printer (Stratasys J750) and post-processed to remove support material, smooth surfaces, and ensure regulatory compliance. The biocompatibility of the materials was assessed according to ISO 10993 standards. The entire process from scanning to digital model creation takes less than 30 minutes, with the 3D printing itself requiring only 6 minutes. The final prosthesis typically undergoes minor adjustments by an ocularist to optimize fit and function.

The automated process was successfully used to create ocular prostheses for 10 patients (4 male, 6 female). The study compared the automated process to traditional manual techniques. The automated process reduced ocularist labor by a factor of five. The ocularist graded the prosthesis shapes and cosmesis. For eight of the ten patients, at least one prosthesis shape could be adjusted to achieve a good fit. For two patients (without orbital implants), the initial shapes were unacceptable, requiring the use of a 3D scan of an existing prosthesis. The cosmesis assessment showed that the automated process achieved excellent or very good results for almost all aspects of appearance (iris size, shape, color; pupil shape; sclera color and veining; overall facial appearance). Patients reported high satisfaction with the realism and color match of the iris. The time required for prosthesis supply and adjustment averaged 60 minutes, encompassing fitting, adjustment, polishing, and assessment.

The findings demonstrate the feasibility and advantages of a fully automated data-driven approach for ocular prosthesis production. The method accurately replicates the color and anatomy of the fellow eye, significantly reducing the labor required by the ocularist and producing highly consistent results. This has potential cost-saving implications when scaled up and the potential for remote prosthesis supply. The use of AS-OCT avoids ionizing radiation, and the digital workflow allows for continuous improvement through data analysis and feedback. While the prostheses typically require minor adjustments by an ocularist, the majority of the manufacturing process is automated. The limitations include the inability to capture highly complex sockets and certain patient conditions (nystagmus, strabismus). The current process is suitable for approximately 80% of patients requiring an ocular prosthesis.

This study presents a significant advancement in ocular prosthesis manufacturing. The automated, data-driven process improves efficiency, reproducibility, and patient outcomes. Future research should focus on expanding the applicability of the method to a broader range of patient needs and exploring the use of advanced materials and 3D printing technologies for even more realistic and durable prostheses. The development could extend to other prosthesis types, such as dental restorations or facial prostheses.

The primary limitations are the need for ocularist adjustments to the final prosthesis shape in most cases and the inability to process patients with highly complex sockets or specific eye conditions such as nystagmus or strabismus. The conformer selection and its placement in the socket may influence the accuracy of the shape prediction. The study had a relatively small sample size, and longer-term clinical studies are needed to fully assess the durability and performance of the 3D-printed prostheses.