No abstract

Artificial intelligence (AI) and machine learning (ML) have been increasingly used in materials science to build predictive models and accelerate discovery. For selected properties, availability of large databases has also facilitated application of deep learning (DL) and transfer learning (TL). However, unavailability of large datasets for a majority of properties prohibits widespread application of DL/TL. We present a cross-property deep-transfer-learning framework that leverages models trained on large datasets to build models on small datasets of different properties. We test the proposed framework on 39 computational and two experimental datasets and find that the TL models with only elemental fractions as input outperform ML/DL models trained from scratch even when they are allowed to use physical attributes as input, for 27/39 (≈ 69%) computational and both the experimental datasets. We believe that the proposed framework can be widely useful to tackle the small data challenge in applying AI/ML in materials science.

Purpose To investigate four-dimensional (4D) flow CT for the assessment of intracardiac blood flow patterns as compared with 4D flow MRI. Materials and Methods This prospective study acquired coronary CT angiography and 4D flow MRI data between February and December 2016 in a cohort of 12 participants (age range, 36-74 years; mean age, 57 years; seven men [age range, 36-74 years; mean age, 57 years] and five women [age range, 52-73 years; mean age, 64 years]). Flow simulations based solely on CT-derived cardiac anatomy were assessed together with 4D flow MRI measurements. Flow patterns, flow rates, stroke volume, kinetic energy, and flow components were quantified for both techniques and were compared by using linear regression. Results Cardiac flow patterns obtained by using 4D flow CT were qualitatively similar to 4D flow MRI measurements, as graded by three independent observers. The Cohen κ score was used to assess intraobserver variability (0.83, 0.79, and 0.70) and a paired Wilcoxon rank-sum test showed no significant change (P > .05) between gradings. Peak flow rate and stroke volumes between 4D flow MRI measurements and 4D flow CT measurements had high correlation (r = 0.98 and r = 0.81, respectively; P < .05 for both). Integrated kinetic energy quantified at peak systole correlated well (r = 0.95, P < .05), while kinetic energy levels at early and late filling showed no correlation. Flow component analysis showed high correlation for the direct and residual components, respectively (r = 0.93, P < .05 and r = 0.87, P < .05), while the retained and delayed components showed no correlation. Conclusion Four-dimensional flow CT produced qualitatively and quantitatively similar intracardiac blood flow patterns compared with the current reference standard, four-dimensional flow MRI. © RSNA, 2018 Online supplemental material is available for this article.

No abstract

Blood flow simulations are making their way into the clinic, and much attention is given to estimation of fractional flow reserve in coronary arteries. Intracardiac blood flow simulations also show promising results, and here the flow field is expected to depend on the pulmonary venous (PV) flow rates. In the absence of in vivo measurements, the distribution of the flow from the individual PVs is often unknown and typically assumed. Here, we performed intracardiac blood flow simulations based on time-resolved computed tomography on three patients, and investigated the effect of the distribution of PV flow rate on the flow field in the left atrium and ventricle. A design-of-experiment approach was used, where PV flow rates were varied in a systematic manner. In total 20 different simulations were performed per patient, and compared to in vivo 4D flow MRI measurements. Results were quantified by kinetic energy, mitral valve velocity profiles and root-mean-square errors of velocity. While large differences in atrial flow were found for varying PV inflow distributions, the effect on ventricular flow was negligible, due to a regularizing effect by mitral valve. Equal flow rate through all PVs most closely resembled in vivo measurements and is recommended in the absence of a priori knowledge.Electronic supplementary materialThe online version of this article (10.1007/s10439-018-02153-5) contains supplementary material, which is available to authorized users.

No abstract

No abstract

The application of machine learning (ML) techniques in materials science has attracted significant attention in recent years, due to their impressive ability to efficiently extract data-driven linkages from various input materials representations to their output properties. While the application of traditional ML techniques has become quite ubiquitous, there have been limited applications of more advanced deep learning (DL) techniques, primarily because big materials datasets are relatively rare. Given the demonstrated potential and advantages of DL and the increasing availability of big materials datasets, it is attractive to go for deeper neural networks in a bid to boost model performance, but in reality, it leads to performance degradation due to the vanishing gradient problem. In this paper, we address the question of how to enable deeper learning for cases where big materials data is available. Here, we present a general deep learning framework based on Individual Residual learning (IRNet) composed of very deep neural networks that can work with any vector-based materials representation as input to build accurate property prediction models. We find that the proposed IRNet models can not only successfully alleviate the vanishing gradient problem and enable deeper learning, but also lead to significantly (up to 47%) better model accuracy as compared to plain deep neural networks and traditional ML techniques for a given input materials representation in the presence of big data.