et al.. A model order reduction approach to create patient-specific mechanical models of human liver in computational medicine applications.. Computer Methods and Programs in Biomedicine, Elsevier, 2019, 170, pp.a b s t r a c tBackground and objective: This paper focuses on computer simulation aspects of Digital Twin models in the medical framework. In particular, it addresses the need of fast and accurate simulators for the mechanical response at tissue and organ scale and the capability of integrating patient-specific anatomy from medical images to pinpoint the individual variations from standard anatomical models. Methods:We propose an automated procedure to create mechanical models of the human liver with patient-specific geometry and real time capabilities. The method hinges on the use of Statistical Shape Analysis to extract the relevant anatomical features from a database of medical images and Model Order Reduction to compute an explicit parametric solution for the mechanical response as a function of such features. The Sparse Subspace Learning, coupled with a Finite Element solver, was chosen to create lowrank solutions using a non-intrusive sparse sampling of the feature space.Results: In the application presented in the paper, the statistical shape model was trained on a database of 385 three dimensional liver shapes, extracted from medical images, in order to create a parametrized representation of the liver anatomy. This parametrization and an additional parameter describing the breathing motion in linear elasticity were then used as input in the reduced order model. Results show a consistent agreement with the high fidelity Finite Element models built from liver images that were excluded from the training dataset. However, we evidence in the discussion the difficulty of having compact shape parametrizations arising from the extreme variability of the shapes found in the dataset and we propose potential strategies to tackle this issue. Conclusions:A method to represent patient-specific real-time liver deformations during breathing is proposed in linear elasticity. Since the proposed method does not require any adaptation to the direct Finite Element solver used in the training phase, the procedure can be easily extended to more complex nonlinear constitutive behaviors -such as hyperelasticity -and more general load cases. Therefore it can be integrated with little intrusiveness to generic simulation software including more sophisticated and realistic models.

The main objective of this study is to combine the statistical shape analysis (SSA) with a morphing procedure in order to generate shape-parametric finite element models of tissues and organs and to explore the reliability and the limitations of this approach when applied to databases of real medical images. As classical statistical shape models are not always adapted to the morphing procedure, a new registration method was developed in order to maximize the morphing efficiency. The method was compared to the traditional iterative thin plate spline (iTPS). Two data sets of 33 proximal femora shapes and 385 liver shapes were used for the comparison. The principal component analysis (PCA) was used to get the principal morphing modes. In terms of anatomical shape reconstruction (evaluated through the criteria of generalization, compactness, and specificity), our approach compared fairly well to the iTPS method, while performing remarkably better in terms of mesh quality, since it was less prone to generate invalid meshes in the interior. This was particularly true in the liver case. Such methodology offers a potential application for the generation of automated finite element (FE) models from medical images. Parametrized anatomical models can also be used to assess the influence of inter-patient variability on the biomechanical response of the tissues. Indeed, thanks to the shape parametrization the user would easily have access to a valid FE model for any shape belonging to the parameters subspace.

The earlier studies on brain vasculature semantic segmentation used classical image analysis methods to extract the vascular tree from images. Nowadays, deep learning methods are widely exploited for various image analysis tasks. One of the strong restrictions when dealing with neural networks in the framework of semantic segmentation is the need to dispose of a ground truth segmentation dataset, on which the task will be learned. It may be cumbersome to manually segment the arteries in a 3D volumes (MRA-TOF typically). In this work, we aim to tackle the vascular tree segmentation from a new perspective. Our objective is to build an image dataset from mouse vasculatures acquired using CT-Scans, and enhance these vasculatures in such a way to precisely mimic the statistical properties of the human brain. The segmentation of mouse images is easily automatized thanks to their specific acquisition modality. Thus, such a framework allows to generate the data necessary for the training of a Convolutional Neural Networki.e. the enhanced mouse images and there corresponding ground truth segmentation -without requiring any manual segmentation procedure. However, in order to generate an image dataset having consistent properties (strong resemblance with MRA images), we have to ensure that the statistical properties of the enhanced mouse images do match correctly the human MRA acquisitions. In this work, we evaluate at length the similarities between the human arteries as acquired on MRA-TOF and the "humanized" mouse arteries produced by our model. Finally, once the model duly validated, we experiment its applicability with a Convolutional Neural Network.

No abstract

No abstract