Improving Consistency in CT Imaging with Deep Learning
Author Information
Author(s): Lee Seul Bi, Hong Youngtaek, Cho Yeon Jin, Jeong Dawun, Lee Jina, Choi Jae Won, Hwang Jae Yeon, Lee Seunghyun, Choi Young Hun, Cheon Jung-Eun, Ji Zhanlin, Zhao Li
Primary Institution: Seoul National University Hospital
Hypothesis
Can deep learning-based image harmonization improve the reproducibility of radiomics features in abdominal CT scans?
Conclusion
Deep learning-based harmonization significantly enhances the reproducibility of radiomics features in abdominal CT images.
Supporting Evidence
- Harmonization improved reproducibility of vessel features from 14% to 69%.
- Reproducible features increased from 18% to 65% after harmonization.
- An average of 179 additional reproducible features per patient was observed post-harmonization.
Takeaway
This study shows that using smart computer programs can help make medical images more consistent, which is important for doctors to make better decisions.
Methodology
The study used a deep learning algorithm to harmonize CT images from different protocols and assessed reproducibility through intraclass correlation coefficients.
Limitations
The external validation was limited to a single type of CT scanner and did not assess reproducibility across true pathologies.
Participant Demographics
117 patients (142 scans) for training and internal validation; 63 patients for external validation.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website