On the use of machine learning techniques for the mechanical characterization of soft biological tissues

Data-driven Modeling of the Mechanical Behavior of Anisotropic Soft Biological Tissue

Closed-form constitutive models are the standard to describe soft tissue mechanical behavior. However, inherent pitfalls of an explicit functional form include poor fits to the data, non-uniqueness of fit, and sensitivity to parameters. Here we design deep neural networks (DNN) that satisfy desirable physics constraints in order to replace expert models of tissue mechanics. To guarantee stress-objectivity, the DNN takes strain (pseudo)-invariants as inputs, and outputs the strain energy and its derivatives. Polyconvexity of strain energy is enforced through the loss function. Direct prediction of both energy and derivative functions enables the computation of the elasticity tensor needed for a finite element implementation. We showcase the DNN ability to learn the anisotropic mechanical behavior of porcine and murine skin from biaxial test data. A multi-fidelity scheme that combines high fidelity experimental data with a low fidelity analytical approximation yields the best performance. Finite element simulations of tissue expansion with the DNN model illustrate the potential of this method to impact medical device design for skin therapeutics. We expect that the open data and software from this work will broaden the use of data-driven constitutive models of tissue mechanics.

Deep Learning for Computational Hemodynamics: A Brief Review of Recent Advances

Computational fluid dynamics (CFD) modeling of blood flow plays an important role in better understanding various medical conditions, designing more effective drug delivery systems, and developing novel diagnostic methods and treatments. However, despite significant advances in computational technology and resources, the expensive computational cost of these simulations still hinders their transformation from a research interest to a clinical tool. This bottleneck is even more severe for image-based, patient-specific CFD simulations with realistic boundary conditions and complex computational domains, which make such simulations excessively expensive. To address this issue, deep learning approaches have been recently explored to accelerate computational hemodynamics simulations. In this study, we review recent efforts to integrate deep learning with CFD and discuss the applications of this approach in solving hemodynamics problems, such as blood flow behavior in aorta and cerebral arteries. We also discuss potential future directions in the field. In this review, we suggest that incorporating physiologic understandings and underlying fluid mechanics laws in deep learning models will soon lead to a paradigm shift in the development novel non-invasive computational medical decisions.

Predicting cell behaviour parameters from glioblastoma on a chip images. A deep learning approach

The broad possibilities offered by microfluidic devices in relation to massive data monitoring and acquisition open the door to the use of deep learning technologies in a very promising field: cell culture monitoring. In this work, we develop a methodology for parameter identification in cell culture from fluorescence images using Convolutional Neural Networks (CNN). We apply this methodology to the in vitro study of glioblastoma (GBM), the most common, aggressive and lethal primary brain tumour. In particular, the aim is to predict the three parameters defining the go or grow GBM behaviour, which is determinant for the tumour prognosis and response to treatment. The data used to train the network are obtained from a mathematical model, previously validated with in vitro experimental results. The resulting CNN provides remarkably accurate predictions (Pearson's ρ > 0.99 for all the parameters). Besides, it proves to be sound, to filter noise and to generalise. After training and validation with synthetic data, we predict the parameters corresponding to a real image of a microfluidic experiment. The obtained results show good performance of the CNN. The proposed technique may set the first steps towards patient-specific tools, able to predict in real-time the tumour evolution for each particular patient, thanks to a combined in vitro-in silico approach.

On the identification of thyroid nodules using semi-supervised deep learning

Detecting malign cases from thyroid nodule examinations is crucial in healthcare particularly to improve the early detection of such cases. However, malign thyroid nodules can be extremely rare and is hard to find using the traditional rule based or expert-based methods. For this reason, the solutions backed by Machine Learning (ML) algorithms are key to improve the detection rates of such rare cases. In this paper, we investigate the application of ML in the healthcare domain for the detection of rare thyroid nodules. The utilized dataset is collected from 636 distinct patients in 99 unique days in Turkey. In addition to the texture feature data of the Ultrasound (US), we have also included the scores of different assessment methods created by different health institutions (e.g., Korean, American and European thyroid societies) as additional features. For detection of extremely rare malign cases, we use auto-encoder based neural network model. Through numerical results, it is shown that the auto-encoder based model can result in an average Recall score of 0.98 and a Sensitivity score of 1.00 for detecting malign and non-malign cases from the healthcare dataset outperforming the traditional classification algorithms that are trained after Synthetic Minority Oversampling Technique (SMOTE) oversampling.

Machine Learning and the Future of Cardiovascular Care: JACC State-of-the-Art Review

The role of physicians has always been to synthesize the data available to them to identify diagnostic patterns that guide treatment and follow response. Today, increasingly sophisticated machine learning algorithms may grow to support clinical experts in some of these tasks. Machine learning has the potential to benefit patients and cardiologists, but only if clinicians take an active role in bringing these new algorithms into practice. The aim of this review is to introduce clinicians who are not data science experts to key concepts in machine learning that will allow them to better understand the field and evaluate new literature and developments. The current published data in machine learning for cardiovascular disease is then summarized, using both a bibliometric survey, with code publicly available to enable similar analysis for any research topic of interest, and select case studies. Finally, several ways that clinicians can and must be involved in this emerging field are presented.

Estimating aortic thoracic aneurysm rupture risk using tension-strain data in physiological pressure range: an in vitro study

Previous studies have shown that the rupture properties of an ascending thoracic aortic aneurysm (ATAA) are strongly correlated with the pre-rupture response features. In this work, we present a two-step machine learning method to predict where the rupture is likely to occur in ATAA and what safety reserve the structure may have. The study was carried out using ATAA specimens from 15 patients who underwent surgical intervention. Through inflation test, full-field deformation data and post-rupture images were collected, from which the wall tension and surface strain distributions were computed. The tension-strain data in the pressure range of 9-18 kPa were fitted to a third-order polynomial to characterize the response properties. It is hypothesized that the region where rupture is prone to initiate is associated with a high level of tension buildup. A machine learning method is devised to predict the peak risk region. The predicted regions were found to match the actual rupture sites in 13 samples out of the total 15. In the second step, another machine learning model is utilized to predict the tissue's rupture strength in the peak risk region. Results suggest that the ATAA rupture risk can be reasonably predicted using tension-strain response in the physiological range. This may open a pathway for evaluating the ATAA rupture propensity using information of in vivo response.

Prediction of local strength of ascending thoracic aortic aneurysms

Knowledges of both local stress and strength are needed for a reliable evaluation of the rupture risk for ascending thoracic aortic aneurysm (ATAA). In this study, machine learning is applied to predict the local strength of ATAA tissues based on tension-strain data collected through in vitro inflation tests on tissue samples. Inputs to machine learning models are tension, strain, slope, and curvature values at two points on the low strain region of the tension-strain curve. The models are trained using data from locations where the tissue ruptured, and subsequently applied to data from intact sites to predict the local rupture strength. The predicted strengths are compared with the known strength at rupture sites as well as the highest tension the tissues experienced at the intact sites. A local rupture index, which is the ratio of the end tension to the predicted rupture strength, is computed. The 'hot spots' of the rupture index are found to match the rupture sites better than those of the peak tension. The study suggests that the strength of ATAA tissue could be reliably predicted from early phase response features defined in this work.

A generic physics-informed neural network-based constitutive model for soft biological tissues

Constitutive modeling is a cornerstone for stress analysis of mechanical behaviors of biological soft tissues. Recently, it has been shown that machine learning (ML) techniques, trained by supervised learning, are powerful in building a direct linkage between input and output, which can be the strain and stress relation in constitutive modeling. In this study, we developed a novel generic physics-informed neural network material (NNMat) model which employs a hierarchical learning strategy by following the steps: (1) establishing constitutive laws to describe general characteristic behaviors of a class of materials; (2) determining constitutive parameters for an individual subject. A novel neural network structure was proposed which has two sets of parameters: (1) a class parameter set for characterizing the general elastic properties; and (2) a subject parameter set (three parameters) for describing individual material response. The trained NNMat model may be directly adopted for a different subject without re-training the class parameters, and only the subject parameters are considered as constitutive parameters. Skip connections are utilized in the neural network to facilitate hierarchical learning. A convexity constraint was imposed to the NNMat model to ensure that the constitutive model is physically relevant. The NNMat model was trained, cross-validated and tested using biaxial testing data of 63 ascending thoracic aortic aneurysm tissue samples, which was compared to expert-constructed models (Holzapfel-Gasser-Ogden, Gasser-Ogden-Holzapfel, and four-fiber families) using the same fitting and testing procedure. Our results demonstrated that the NNMat model has a significantly better performance in both fitting (R² value of 0.9632 vs 0.9019, p=0.0053) and testing (R² value of 0.9471 vs 0.8556, p=0.0203) than the Holzapfel-Gasser-Ogden model. The proposed NNMat model provides a convenient and general methodology for constitutive modeling.

QSAR without borders

Prediction of chemical bioactivity and physical properties has been one of the most important applications of statistical and more recently, machine learning and artificial intelligence methods in chemical sciences. This field of research, broadly known as quantitative structure-activity relationships (QSAR) modeling, has developed many important algorithms and has found a broad range of applications in physical organic and medicinal chemistry in the past 55+ years. This Perspective summarizes recent technological advances in QSAR modeling but it also highlights the applicability of algorithms, modeling methods, and validation practices developed in QSAR to a wide range of research areas outside of traditional QSAR boundaries including synthesis planning, nanotechnology, materials science, biomaterials, and clinical informatics. As modern research methods generate rapidly increasing amounts of data, the knowledge of robust data-driven modelling methods professed within the QSAR field can become essential for scientists working both within and outside of chemical research. We hope that this contribution highlighting the generalizable components of QSAR modeling will serve to address this challenge.

Augmented patient-specific functional medical imaging by implicit manifold learning

This paper uses machine learning to enrich magnetic resonance angiography and magnetic resonance imaging acquisitions. A convolutional neural network is built and trained over a synthetic database linking geometrical parameters and mechanical characteristics of the arteries to blood flow rates and pressures in an arterial network. Once properly trained, the resulting neural network can be used in order to predict blood pressure in cerebral arteries noninvasively in nearly real-time. One challenge here is that not all input variables present in the synthetic database are known from patient-specific medical data. To overcome this challenge, a learning technique, which we refer to as implicit manifold learning, is employed: in this view, the input and output data of the neural network are selected based on their availability from medical measurements rather than being defined from the mechanical description of the arterial system. The results show the potential of the method and that machine learning is an alternative to costly ensemble based inversion involving sophisticated fluid structure models.

A feasibility study of deep learning for predicting hemodynamics of human thoracic aorta

Numerical analysis methods including finite element analysis (FEA), computational fluid dynamics (CFD), and fluid-structure interaction (FSI) analysis have been used to study the biomechanics of human tissues and organs, as well as tissue-medical device interactions, and treatment strategies. However, for patient-specific computational analysis, complex procedures are usually required to set-up the models, and long computing time is needed to perform the simulation, preventing fast feedback to clinicians in time-sensitive clinical applications. In this study, by using machine learning techniques, we developed deep neural networks (DNNs) to directly estimate the steady-state distributions of pressure and flow velocity inside the thoracic aorta. After training on hemodynamic data from CFD simulations, the DNNs take as input a shape of the aorta and directly output the hemodynamic distributions in one second. The trained DNNs are capable of predicting the velocity magnitude field with an average error of 1.9608% and the pressure field with an average error of 1.4269%. This study demonstrates the feasibility and great potential of using DNNs as a fast and accurate surrogate model for hemodynamic analysis of large blood vessels.

Engineering approaches for characterizing soft tissue mechanical properties: A review

From cancer diagnosis to detailed characterization of arterial wall biomechanics, the elastic property of tissues is widely studied as an early sign of disease onset. The fibrous structural features of tissues are a direct measure of its health and functionality. Alterations in the structural features of tissues are often manifested as local stiffening and are early signs for diagnosing a disease. These elastic properties are measured ex vivo in conventional mechanical testing regimes, however, the heterogeneous microstructure of tissues can be accurately resolved over relatively smaller length scales with enhanced spatial resolution using techniques such as micro-indentation, microelectromechanical (MEMS) based cantilever sensors and optical catheters which also facilitate in vivo assessment of mechanical properties. In this review, we describe several probing strategies (qualitative and quantitative) based on the spatial scale of mechanical assessment and also discuss the potential use of machine learning techniques to compute the mechanical properties of soft tissues. This work details state of the art advancement in probing strategies, associated challenges toward quantitative characterization of tissue biomechanics both from an engineering and clinical standpoint.