Dental artifact corruption classifier for Head and Neck CT images
Keywords: Computed Tomography, Artifact detection, Autoencoders, Convolutional Neural Networks, Oral cavity
TL;DR: We demonstrate an efficient method to classify dental artifact in Head and Neck CT images.
Abstract: Image artifacts emanating from dental implants can inappropriately bias the training of machine learning models for segmentation. To ameliorate the corruption of the segmentation tool, we developed and tested a dental artifact classifier to grade 2D oral cavity images as having no detectable, moderate, or severe artifact. A more balanced training dataset was selected by constraining the artifact classifier to the oral cavity region. This was achieved by applying an autoencoder, which was trained only on oral cavity images, to the entire stack to determine whether an image was in the oral cavity. Images with low reconstruction error were classified as oral cavity and input to a multi-class 2D convolutional neural network for the grade of an artifact. The classification was then written back into the DICOM metadata so that it may be reliably selected for subsequent usage based on artifact status. This type of approach may be applied to quality control of training datasets from uncurated sources, such as publicly available collections or de-identified patient data.
Paper Type: methodological development
Primary Subject Area: Detection and Diagnosis
Secondary Subject Area: Segmentation
Paper Status: original work, not submitted yet
Source Code Url: https://github.com/prashulsingh/CT_Dental_Artifact_Classifier
Data Set Url: https://wiki.cancerimagingarchive.net/pages/viewpage.action?pageId=33948764
Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.
Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.
4 Replies
Loading