Anatomically Embedded Representations for Medical Image Registration in Clinical Applications

医学图像配准是诊断、病理随访、治疗计划和模拟、术中手术辅助和治疗结果评估等多种临床应用中的一项重要任务。尽管研究界已经取得了巨大的进展，提出了大量的复杂的方法内模态和多模态图像配准，其在临床环境中的适用性仍然是有限的。阻止这些方法在临床应用中广泛使用的最重要的瓶颈不仅是它们的精度和鲁棒性，而且是它们相关的计算成本。我们的研究项目的总体目标是进行基础和应用研究，以开发躯干解剖结构的新表示和配准算法，这些算法在青少年特发性脊柱侧凸畸形的治疗中对非侵入性诊断，临床随访，手术计划和辅助具有很高的潜在影响。更具体地说，我们将煽动一个新的表示的三角网格表面的树干，从非侵入性的3D相机，使用谱图理论。我们的方法的新奇之处在于使用躯干的解剖特征作为图节点的权重和额外的嵌入坐标，而不需要额外的计算费用。此外，通过使用基于脊柱关节模型的闭合形式解决方案，将提高躯干的多模态图像配准的效率。该研究计划将首先通过将解剖学相关表示与原始谱嵌入方法相结合，为医学图像配准领域的科学研究进步做出重大贡献。其次，所提出的方法将确保临床应用中所需的多模态图像配准的再现性和效率，这将有助于医疗保健的技术转让。第三，该研究项目促进了无创模式在脊柱侧凸诊断、预后和治疗结果评估中的应用，这将提高患者的生活质量，降低医疗保健系统的成本。

Medical image registration is an essential task in a variety of clinical applications such as diagnosis, follow-up of pathologies, treatment planning and simulation, intra-operative surgery assistance and treatment outcome assessment. Despite the tremendous progress already made by the research community and the huge number of sophisticated methods proposed for intra-modal and multimodal image registration, their applicability in clinical environments is still limited. The most important bottlenecks preventing a wide use of these methods in clinical applications are not only their precision and robustness but also their associated computational cost. The overall objective of our research program is to pursue fundamental and applied research for the development of novel representations of anatomical structures of the trunk and registration algorithms with high potential impact in non invasive diagnosis, clinical follow-up, surgical planning and assistance in the treatment of adolescent idiopathic scoliosis deformities. More specifically, we will instigate a new representation of the triangulated mesh surface of the trunk, obtained from non invasive 3D cameras, using spectral graph theory. The novelty of our approach will lie in using anatomical features of the trunk as weights in graph nodes and as extra embedded coordinates with no additional computational expense. In addition, the efficiency of multimodal image registration of the trunk will be enhanced by using a closed form solution based on an articulated model of the spine. This research program will first significantly contribute to the advancement of scientific research in the field of medical image registration by combining anatomically relevant representations with an original spectral embedding approach. Second, the proposed approach will ensure the reproducibility and the efficiency of multimodal image registration required in clinical applications, which will contribute toward technology transfer for healthcare. Third, this research program promotes the use of non invasive modalities in the diagnosis, prognosis and treatment outcome assessment of scoliosis, which will lead to enhanced quality of life for patients and reduced costs for the healthcare system.