Self-Supervised Learning and 3D Printing Technology in Facial Reconstruction and Defect Coverage

Abstract

This study proposes a model for creating facial wound-covering masks to support patients recovering from injuries, especially those with scars or deformities resulting from accidents and wars. The model aims to increase patient confidence, improve wound hygiene, and protect the environment. A novel dataset was developed, consisting of diverse facial images with various scar conditions simulated from real human scars. The study employs self-supervised learning (SSL) with a pretrained base model to convert 2D images into 3D representations without compromising critical facial features. SSL is implemented during the encoding phase, allowing the model to familiarize itself with new data. Through the integration of 3D printing technology, the entire process, from wound reconstruction to product manufacturing, has been tested in the laboratory. The results indicate that the model not only effectively covers wounds but also restores the original facial structure nearly perfectly. The improvement is clearly demonstrated through error reduction and increased accuracy across experiments with diverse datasets. This research opens new possibilities for practical applications, particularly for war victims, by offering a novel, safe, and convenient treatment solution.