SAACT: Semiautomated Annotation of Computerized Tomography Data
- 14 October 2025
- journal article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Canadian Journal of Electrical and Computer Engineering
- Vol. 48 (4) , 370-380
Abstract
A successful deep learning-based solution design requires a large volume of well-annotated data to ensure model generalizability and efficient deployment. For certain advanced applications, such as semantic segmentation, the training dataset must be manually annotated by assigning labels to each pixel in the images. This labor-intensive and time-consuming process must be performed and verified by domain experts. This article presents a semiautomated data annotation technique for X-ray computed tomography (XCT) data, leveraging computer-aided design (CAD) design files. The proposed system employs various preprocessing techniques, including noise filtering and background removal. Additionally, we introduce an improved 3-D volume registration method based on the diffusion imaging in python (DIPY) library. The proposed annotation framework was applied to both real and semantic XCT datasets for an industrial tool and validated using a semantic segmentation model. The trained model achieved intersection over union (IoU) scores of 0.70 and 0.64 for the real and semantic XCT data, respectively. These results demonstrate the effectiveness of the annotation method, indicating strong performance in both cases. The findings confirm that the framework can be integrated into artificial intelligence (AI)-based industrial inspection systems to accelerate the industrial inspection processes, improve defect detection accuracy, and enable automated report generation.Keywords
Funding Information
- New Vision Systems Canada Inc. (NVS) and Mitacs
This publication has 31 references indexed in Scilit:
- Rapid data annotation for sand-like granular instance segmentation using mask-RCNNAutomation in Construction, 2022
- Biological data annotation via a human-augmenting AI-based labeling systemnpj Digital Medicine, 2021
- Defect detection in CT scans of cast aluminum parts: A machine vision perspectiveNeurocomputing, 2021
- nnU-Net: a self-configuring method for deep learning-based biomedical image segmentationNature Methods, 2020
- Lattice-Boltzmann interactive blood flow simulation pipelineInternational Journal of Computer Assisted Radiology and Surgery, 2020
- Deep learning in medical image registration: a surveyMachine Vision and Applications, 2020
- Data augmentation using generative adversarial networks (CycleGAN) to improve generalizability in CT segmentation tasksScientific Reports, 2019
- Dermatologist-level classification of skin cancer with deep neural networksNature, 2017
- U-Net: Convolutional Networks for Biomedical Image Segmentation2015
- 3D Slicer as an image computing platform for the Quantitative Imaging NetworkMagnetic Resonance Imaging, 2012