Infant Brain Tumor Detection Using Ultra Sound Images and MRI Images

Deformable registration plays an important role in both longitudinal and population-based medical image analysis. Accurately registering infant brain MRI scans is difficult because the brain undergoes rapid structural growth and contrast variations during early development. This challenge affects both longitudinal scans of the same infant and cross-sectional scans across different infants. To address this issue, this paper presents a reusable recurrent deep learning framework for infant brain MRI registration. The proposed approach has several important features. First, instead of relying directly on image intensity values, the method utilizes segmented brain tissue maps for registration, which helps reduce problems caused by rapid intensity and contrast changes during infancy. The manuscript was initially submitted on November 3, 2021, revised on December 11, 2021, and officially accepted on December 15, 2021. It was published online on December 21, 2021, with the latest version released on May 2, 2022. Research contributions from Gang Li received partial funding from the United States National Institutes of Health (NIH) through Grants MH116225 and MH117943. Li Wang’s work was also partially funded by NIH Grant MH117943. Pew-Thian Yap obtained support from NIH Grant AG053867, while Qian Wang’s contribution was supported by the Science and Technology Commission of Shanghai Municipality (STCSM) under Grant 19QC1400600. The corresponding authors for this work are Dinggang Shen and Qian Wang. This research included studies involving human participants or animals. All experimental methods, ethical procedures, and research protocols were reviewed and approved by the Institutional Review Board (IRB) of the School of Medicine at the University of North Carolina at Chapel Hill. Dongming Wei is with the School of Biomedical Engineering, Shang- hai Jiao Tong University, Shanghai 200030, China, and also with the Department of Radiology and Biomedical Research Imaging Center (BRIC), University of North Carolina, Chapel Hill, NC 27599 USA (e-mail: [email protected]). The text you provided is mainly author affiliation and contact information from a research paper. Such details are factual and standard in academic publications, so they usually do not count as plagiarism when cited properly. However, if you want to avoid direct copying, you can rewrite it like this: Sahar Ahmad, Lei Ma, Zhengwang Wu, Gang Li, Li Wang, Weili Lin, and Pew-Thian Yap are affiliated with the Department of Radiology and the Biomedical Research Imaging Center (BRIC) at the University of North Carolina at Chapel Hill, located in Chapel Hill, North Carolina, USA. Their professional contact information is available through the university’s official communication channels. Yuyu Guo and Liyun Chen are with the School of Biomedical Engineer- ing, Shanghai Jiao Tong University, Shanghai 200030, China (e-mail: [email protected]; [email protected]). Yunzhi Huang is with the College of Biomedical Engineering, Sichuan University, Chengdu 610065, China, and also with the Department of Radiology and Biomedical Research Imaging Center (BRIC), Uni- versity of North Carolina, Chapel Hill, NC 27599 USA (e-mail: [email protected]). Dinggang Shen is associated with the School of Biomedical Engineering at ShanghaiTech University, Shanghai, China. He is also connected with Shanghai United Imaging Intelligence Co., Ltd.. His professional correspondence can be made through his academic email address. Qian Wang is with the School of Biomedical Engineering, ShanghaiTech University, Shanghai 201210, China (e-mail: [email protected]). The proposed approach is designed to handle the rapid developmental changes in infant brain tissues that occur during the first year of life. A single deep registration network is trained using a one-time learning strategy and is repeatedly utilized during the inference stage to gradually estimate complex deformation fields in a step- by-step manner. In addition, the framework introduces an adaptive smoothing mechanism along with a tissue- sensitive anti-folding constraint to maintain biologically realistic deformations while preserving high registration performance. Experimental comparisons with existing state-of-the-art registration techniques demonstrate that the proposed model provides superior registration accuracy and generates smoother deformation fields. The implementation details and source code of the network are publicly accessible through GitHub Repository for ACTA- Reg-Net.