Page 261 - v11i4

International Journal of Bioprinting Deep learning-based 3D digital model of fetal heart

support structures, its impact on the quality of printed fetal labor costs associated with reliable model construction,
heart models is significant. By producing anatomically paving the way for the future development of real-time
accurate and clean digital reconstructions with minimal analysis systems based on 3D digital models, as opposed
noise and well-defined boundaries, FRT substantially to the current 2D anatomical section-based approach.
reduces the need for manual mesh correction and improves Future research efforts will focus on achieving real-time
the fidelity of 3D-printed outputs. In this aspect, the deep analysis of 3D fetal hearts, further enhancing the field of
learning pipeline acts as an upstream quality enhancer prenatal healthcare.
for additive manufacturing. Therefore, future studies may
explore extending AI involvement to the printing process 5. Conclusion
itself—such as adaptive slicing strategies or automated In this study, a fetal heart 3D digital model was constructed
printability evaluation—to further improve the efficiency using FRT based on US volume data. The model may be
and accuracy of model production. used to enhance the clinical diagnosis and treatment of
In terms of practical workflow efficiency, FRT CHD during pregnancy. Our results indicate that deep
significantly outperforms traditional manual methods. learning has the ability to process US data accurately,
Manual reconstruction of a fetal heart digital model representing an important step towards reconstructing
typically requires clinicians to segment US volume data fetal heart digital model and advancing clinical diagnosis
layer by layer from 2D slices—a labor-intensive and time- and treatment of CHD during pregnancy.
consuming process. With FRT, most segmentation tasks are
automated, while clinicians can interactively calibrate the Acknowledgments
output using threshold adjustments. This semi-automatic None.
pipeline reduces the reconstruction time for a single
digital heart cavity model from approximately 5 h to just Funding
5 min. Even without human interaction, the process can be This work was supported by the Key Research and
completed in 2 min, albeit with slightly reduced accuracy. Development Project of Shaanxi Province (2021LLRH-08).
These findings demonstrate the considerable clinical value
of FRT, particularly in reducing workload and enabling Conflict of interest
timely decision-making in prenatal care.
The authors declare that they have no competing financial
Nonetheless, FRT does have some limitations.
Firstly, the size and quality of the dataset have a direct interests or personal relationships that could have appeared
to influence the work reported in this paper.
impact on the accuracy of the deep learning model. The
dataset used in FRT comprises 4852 data streams from Author contributions
100 unique individuals, which could potentially limit
the model’s performance. To address this challenge, Conceptualization: Lijun Yuan, Airong Qian
interactive segmentation that incorporates user Data Curation: Zekai Zhang, Zhuojun Mao
knowledge is integrated, resulting in a more robust Investigation: Wenjuan Zhang, Linbin Lai
segmentation performance. Secondly, while FRT Methodology: Jiahe Liang, Zewen Zhang
significantly reduces the time required for digital model Resources: Yitong Guo, Na Hou
reconstruction, it is not yet capable of achieving real- Supervision: Tiesheng Cao, Yu Li, Lijun Yuan, Airong Qian
time analysis of fetal heart functions. Several limitations Writing–original draft: Wenjuan Zhang, Linbin Lai
contribute to this challenge: (i) obtaining a more robust Writing–review & editing: Tiesheng Cao, Yu Li
performance through the interactive method is time-
consuming; (ii) the low quality of US images and the high Ethics approval and consent to participate
variation between different operators make it difficult This study involved human subjects and was conducted
to extract useful features, limiting the performance of in accordance with the ethical principles outlined in
end-to-end DNNs; and (iii) the limited training data the Declaration of Helsinki. The research protocol
available represents a bottleneck for the application of titled “Three-dimensional Fast Reconstruction in
deep learning methods in prenatal US image analysis. In Fetal Ultrasound Imaging Using Artificial Intelligence
summary, the results of our research represent a crucial Techniques and Three-dimensional (3D) Bioprinting” was
advancement in the reconstruction of digital fetal reviewed and approved by the Institutional Review Board
heart models using US volume data and deep learning (IRB) of Tangdu Hospital, Air Force Medical University
technology. FRT significantly reduces the time and (approval no.: TDLL-202402-01; date: January 5, 2024).

Volume 11 Issue 4 (2025) 253 doi: 10.36922/IJB025200192