Utilising Artificial Intelligence (AI) to Automate Defacing of the Nose in Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) images

Abstract

Purpose: Volume reconstructed CT and MRI head and brain images can be recognized visually, which may conflict with the Personal Data Protection Act when using these images for research purposes. To allow for proper de-identification and to make the face unrecognizable, certain facial features such as the nose, need to be removed. Defacing is one method that can be used to do so. In defacing, facial features are first identified on the images and then removed. The aim of this research project is to use AI to carry out the defacing or removal of the nose in CT and MRI images of the head and brain, while keeping other parts of the anatomy intact so that these images can be de-identified successfully and still be used for various research purposes. Methods: The objective is to train two AI algorithms to deface the noses in CT and MRI images respectively, in an effective and reproducible manner, and to address the limitations of current available programs. To do so, the researchers first located and annotated the nose on the CT and MRI head images. Experts verified that the nose was annotated accurately. These annotated images were then used to train the AI to recognize and locate the nose. Mean average precision (mAP), a metric commonly used to measure the accuracy of AI object detection, was used to gauge the efficacy and success of the AI training. Results: Based on the results of the mAP values, the AI successfully located the nose. Conclusion: In conclusion, utilizing AI, the researchers successfully created two effective and reproducible methods to deface the nose in CT and MRI images respectively. To complete the defacing process for an entire face, more research can be done on utilizing a similar method to remove the other facial features like the eyes and the mouth.