A Real-World Clinical Implementation of Automated Processing Using Intelligent Work Aid for Rapid Reformation at the Orbitomeatal Line in Head Computed Tomography

Abstract

Objectives: The aim was to investigate the time savings and plane accuracy of multivendor head computed tomography (CT) using the intelligent work aid with automatic reformatting of the axial head image at the orbitomeatal line. Materials and Methods: We retrospectively reviewed 781 head CTs (median, 70 years; 441 men) collected by CT systems from 3 vendors. In addition to the orbitomeatal line image reformatted by a CT specialist as a reference, we obtained the fully automated orbitomeatal line image using the intelligent work aid. We calculated the offset angle from the reference of the automatically reformatted image. We defined the large offset angle groups as those with an offset angle greater than 3 degrees. Multivariate logistic regression was used to determine the independent factors for the large offset angle groups. We compared the postprocessing times measured using the intelligent work aid or by a CT specialist. Results: With the intelligent work aid, 99.7% of CTs were automatically reformatted to the orbitomeatal line without error. Furthermore, 88.1% of CTs were within the 3 degrees' offset angle when compared with the reference produced by a CT specialist. The median offset angle from the reference was 1.41 degrees. Multivariate analysis showed that the offset angle of the positioning plane was an independent factor (odds ratio, 1.045; P = 0.005) for predicting the large offset angle group. Furthermore, this technique was 4 times faster (6.4 ± 0.7 seconds) than a CT specialist (25.6 ± 6.4 seconds). Conclusions: The intelligent work aid can generate a fast and precise head CT image aligned at the orbitomeatal line, even in real-world clinical CTs. However, precise positioning remains essential.

Generative adversarial network-based post-processed image super-resolution technology for accelerating brain MRI: comparison with compressed sensing

“…Verification using real-world data is another problem, which is considered necessary for implementing artificial intelligence in medical imaging (26,32). As for future tasks, to thoroughly evaluate the generalization performance, it must include input images that are different from the training data, such as from various vendors and disease backgrounds (33). Further investigation is also needed before the implementation of recently reported state-of-the-art deep-learning technologies (31)(32)(33) in clinical practice.…”

“…As for future tasks, to thoroughly evaluate the generalization performance, it must include input images that are different from the training data, such as from various vendors and disease backgrounds (33). Further investigation is also needed before the implementation of recently reported state-of-the-art deep-learning technologies (31)(32)(33) in clinical practice. The present study has some limitations.…”

Deep learning for lung cancer on computed tomography: early detection and prognostic prediction

“…Deep learning has emerged as a vital player in fields such as object detection, medical image segmentation [9] [10] [11], diagnostics [12] [13], and predictive analytics [14] [15] [16]. Previous research proposed a semi-automatic multiplanar reconstruction method [17]. This method mandates manually setting five head landmarks on axial images to identify the orbitomeatal line, facilitating 3D and multiplanar reconstruction.…”