Searcharxiv⌕ Search

arXiv subjects

Dazhong Wu

Publications and source records attributed to Dazhong Wu.

3 recordsLinked to original sources

Automated Left Ventricle Strain Analysis from DENSE MRI

Background: Quantifying myocardial strains from Displacement ENcoding with Stimulated Echoes (DENSE) MRI provides insight into left ventricular performance, but current data processing pipelines remain limited in automation, robustness, and full left ventricle coverage. Methods: We present a fully automated framework for voxel-wise displacement and strain analysis of the left ventricle from DENSE MRI. Our method integrates a multi-channel nnU-Net model trained on both magnitude and phase images for segmentation, an automated phase unwrapping approach, an adaptation of the regularized spatiotemporal least squares (RSTLS) formulation for displacement computation, and strain calculations based on radial basis function interpolation. The proposed framework enables the estimation of displacements and strains across the left ventricle from a full stack of short-axis images. Results: Validation using a computational phantom shows the framework's overall resilience to data noise. Following computational phantom evaluation, the proposed framework is applied to analyze DENSE MRI short-axis slices acquired in forty-seven (N=47) healthy volunteers and covering the full left ventricle. Circumferential, longitudinal, and radial strains are presented throughout the cardiac cycle and at peak systole, showcasing the performance of the proposed framework. Results show that circumferential and longitudinal strains are computed accurately. However, challenges persist when estimating radial strain, especially in the subendocardial layer. A group-wise strain comparison is also performed based on volunteers' sex and age showing a significant effect of sex on circumferential and longitudinal strains. Conclusions: Overall, the proposed framework provides an automated and modular pipeline to compute displacements and strains from DENSE MR images that can be scaled and generalized toward clinical applications.

q-bio.QM↗

A Dynamic Time Warping-Transfer Learning Approach to Transferring Knowledge in Stress-strain Behaviors from Polymers to Metals: An Affordable and Generalizable Additive Manufacturing Part Qualification Framework

Part qualification in additive manufacturing (AM) ensures that additively manufactured parts can be consistently produced and reliably used in critical applications. One crucial aspect of part qualification is to determine the complex stress-strain behavior of additively manufactured parts. However, conventional part qualification techniques such as the destructive testing and non-destructive testing are costly and time consuming, especially for metal AM. To address this challenge, we develop a dynamic time warping (DTW)-transfer learning (TL) framework for AM part qualification by transferring knowledge gained from the stress-strain behaviors of additively manufactured low-cost polymers to high-performance, expensive metals. Specifically, the framework selects one single optimal polymer dataset that is the most similar to the metal dataset in the target domain using DTW among multiple polymer datasets, including Nylon, PLA, CF-ABS, and Resin. A long short-term memory (LSTM) model is then trained on one single optimal polymer dataset and tested on one of three target metal datasets, including AlSi10Mg, Ti6Al4V, and carbon steel datasets. Experimental results show that the Resin dataset is selected as the optimal polymer dataset in the source domain for the AlSi10Mg and Ti6Al4V datasets, while the Nylon dataset is selected as the optimal polymer dataset in the source domain for the carbon steel dataset. The DTWTL model trained on one single optimal polymer dataset as the source domain achieves the best predictive performance, including an average mean absolute percentage error of 12.41%, an average root mean squared error of 63.75, and an average coefficient of determination of 0.96 when three metals are used as the target domain, outperforming the vanilla LSTM model without TL as well as the TL model trained on all four polymer datasets as the source domain.

cs.LG↗

Predicting Stress-strain Behaviors of Additively Manufactured Materials via Loss-based and Activation-based Physics-informed Machine Learning

Predicting the stress-strain behaviors of additively manufactured materials is crucial for part qualification in additive manufacturing (AM). Conventional physics-based constitutive models often oversimplify material properties, while data-driven machine learning (ML) models often lack physical consistency and interpretability. To address these issues, we propose a physics-informed machine learning (PIML) framework to improve the predictive performance and physical consistency for predicting the stress-strain curves of additively manufactured polymers and metals. A polynomial regression model is used to predict the yield point from AM process parameters, then stress-strain curves are segmented into elastic and plastic regions. Two long short-term memory (LSTM) models are trained to predict two regions separately. For the elastic region, Hooke's law is embedded into the LSTM model for both polymer and metal. For the plastic region, Voce hardening law and Hollomon's law are embedded into the LSTM model for polymer and metal, respectively. The loss-based and activation-based PIML architectures are developed by embedding the physical laws into the loss and activation functions, respectively. The performance of the two PIML architectures are compared with two LSTM-based ML models, three additional ML models, and a physics-based constitutive model. These models are built on experimental data collected from two additively manufactured polymers (i.e., Nylon and carbon fiber-acrylonitrile butadiene styrene) and two additively manufactured metals (i.e., AlSi10Mg and Ti6Al4V). Experimental results demonstrate that two PIML architectures consistently outperform the other models. The segmental predictive model with activation-based PIML architecture achieves the lowest MAPE of 10.46+/-0.81% and the highest R^2 of 0.82+/-0.05 arocss four datasets.

cs.LG↗