Abstract:Bubble segmentation and size detection algorithms have been developed in recent years for their high efficiency and accuracy in measuring bubbly two-phase flows. In this work, we proposed an architecture called bubble generative adversarial networks (BubGAN) for the generation of realistic synthetic images which could be further used as training or benchmarking data for the development of advanced image processing algorithms. The BubGAN is trained initially on a labeled bubble dataset consisting of ten thousan… Show more
“…one out of five images was generated inaccurately). GANs are very suitable for expanding the variability of the training database where all variations of objects are allowed [24]. The method has shown its superiority in generating data for medical imaging in solving unsupervised classification problem, which suffers from a small training set and includes only two classes of images (i.e.…”
BackgroundInfertility and subfertility affect a significant proportion of humanity. Assisted reproductive technology has been proven capable of alleviating infertility issues. In vitro fertilisation is one such option whose success is highly dependent on the selection of a high-quality embryo for transfer. This is typically done manually by analysing embryos under a microscope. However, evidence has shown that the success rate of manual selection remains low. The use of new incubators with integrated time-lapse imaging system is providing new possibilities for embryo assessment. As such, we address this problem by proposing an approach based on deep learning for automated embryo quality evaluation through the analysis of time-lapse images. Automatic embryo detection is complicated by the topological changes of a tracked object. Moreover, the algorithm should process a large number of image files of different qualities in a reasonable amount of time.MethodsWe propose an automated approach to detect human embryo development stages during incubation and to highlight embryos with abnormal behaviour by focusing on five different stages. This method encompasses two major steps. First, the location of an embryo in the image is detected by employing a Haar feature-based cascade classifier and leveraging the radiating lines. Then, a multi-class prediction model is developed to identify a total cell number in the embryo using the technique of deep learning.ResultsThe experimental results demonstrate that the proposed method achieves an accuracy of at least 90% in the detection of embryo location. The implemented deep learning approach to identify the early stages of embryo development resulted in an overall accuracy of over 92% using the selected architectures of convolutional neural networks. The most problematic stage was the 3-cell stage, presumably due to its short duration during development.ConclusionThis research contributes to the field by proposing a model to automate the monitoring of early-stage human embryo development. Unlike in other imaging fields, only a few published attempts have involved leveraging deep learning in this field. Therefore, the approach presented in this study could be used in the creation of novel algorithms integrated into the assisted reproductive technology used by embryologists.
“…one out of five images was generated inaccurately). GANs are very suitable for expanding the variability of the training database where all variations of objects are allowed [24]. The method has shown its superiority in generating data for medical imaging in solving unsupervised classification problem, which suffers from a small training set and includes only two classes of images (i.e.…”
BackgroundInfertility and subfertility affect a significant proportion of humanity. Assisted reproductive technology has been proven capable of alleviating infertility issues. In vitro fertilisation is one such option whose success is highly dependent on the selection of a high-quality embryo for transfer. This is typically done manually by analysing embryos under a microscope. However, evidence has shown that the success rate of manual selection remains low. The use of new incubators with integrated time-lapse imaging system is providing new possibilities for embryo assessment. As such, we address this problem by proposing an approach based on deep learning for automated embryo quality evaluation through the analysis of time-lapse images. Automatic embryo detection is complicated by the topological changes of a tracked object. Moreover, the algorithm should process a large number of image files of different qualities in a reasonable amount of time.MethodsWe propose an automated approach to detect human embryo development stages during incubation and to highlight embryos with abnormal behaviour by focusing on five different stages. This method encompasses two major steps. First, the location of an embryo in the image is detected by employing a Haar feature-based cascade classifier and leveraging the radiating lines. Then, a multi-class prediction model is developed to identify a total cell number in the embryo using the technique of deep learning.ResultsThe experimental results demonstrate that the proposed method achieves an accuracy of at least 90% in the detection of embryo location. The implemented deep learning approach to identify the early stages of embryo development resulted in an overall accuracy of over 92% using the selected architectures of convolutional neural networks. The most problematic stage was the 3-cell stage, presumably due to its short duration during development.ConclusionThis research contributes to the field by proposing a model to automate the monitoring of early-stage human embryo development. Unlike in other imaging fields, only a few published attempts have involved leveraging deep learning in this field. Therefore, the approach presented in this study could be used in the creation of novel algorithms integrated into the assisted reproductive technology used by embryologists.
“…For a training dataset, we used both experimental and synthetic bubbly flow images obtained from the upward bubbly flows in an expansion pipe 8 and BubGAN algorithm 40 , respectively (Fig. 1 a,b).…”
Section: Training and Evaluation Of The Algorithmmentioning
confidence: 99%
“…To add bubbles smaller than the average size of 35 pixels in diameter, on the other hand, the height of the image with IoU B = 0.16 was adjusted to three times longer than for other cases (IoU B = 0.11 and 0.2), because all training inputs are scaled to be the same size (640 × 640 pixels), regardless of the physical size of the image. Figure 1 Examples of bubbly flow images used as a training and test dataset: ( a ) experimental data of upward bubbly flows in an expansion pipe 8 ; ( b ) synthetic bubble images from BubGAN 40 ; ( c ) experimental data of bubble-swarm flow 9 . …”
Section: Training and Evaluation Of The Algorithmmentioning
confidence: 99%
“… Examples of bubbly flow images used as a training and test dataset: ( a ) experimental data of upward bubbly flows in an expansion pipe 8 ; ( b ) synthetic bubble images from BubGAN 40 ; ( c ) experimental data of bubble-swarm flow 9 . …”
Section: Training and Evaluation Of The Algorithmmentioning
While investigating multiphase flows experimentally, the spatiotemporal variation in the interfacial shape between different phases must be measured to analyze the transport phenomena. For this, numerous image processing techniques have been proposed, showing good performance. However, they require trial-and-error optimization of thresholding parameters, which are not universal for all experimental conditions; thus, their accuracy is highly dependent on human experience, and the overall processing cost is high. Motivated by the remarkable improvements in deep learning-based image processing, we trained the Mask R-CNN to develop an automated bubble detection and mask extraction tool that works universally in gas–liquid two-phase flows. The training dataset was rigorously optimized to improve the model performance and delay overfitting with a finite amount of data. The range of detectable bubble size (particularly smaller bubbles) could be extended using a customized weighted loss function. Validation with different bubbly flows yields promising results, with AP50reaching 98%. Even while testing with bubble-swarm flows not included in the training set, the model detects more than 95% of the bubbles, which is equivalent or superior to conventional image processing methods. The pure processing speed for mask extraction is more than twice as fast as conventional approaches, even without counting the time required for tedious threshold parameter tuning. The present bubble detection and mask extraction tool is available online (https://github.com/ywflow/BubMask).
“…Many CNN-based deep learning frameworks are effective because CNNs emulate the human brain's natural visual perception mechanism by systematically learning features through multiple operational layers 45 . Image-based deep learning models can play a vital role in fully understanding boiling physics because boiling images are richly embedded with bubble statistics, which are quantitative measurements of the dynamic boiling phenomena [46][47][48] . Despite the potential for understanding image-based boiling physics via deep learning frameworks, very few attempts have been made to build them.…”
Boiling is arguably Nature’s most effective thermal management mechanism that cools submersed matter through bubble-induced advective transport. Central to the boiling process is the development of bubbles. Connecting boiling physics with bubble dynamics is an important, yet daunting challenge because of the intrinsically complex and high dimensional of bubble dynamics. Here, we introduce a data-driven learning framework that correlates high-quality imaging on dynamic bubbles with associated boiling curves. The framework leverages cutting-edge deep learning models including convolutional neural networks and object detection algorithms to automatically extract both hierarchical and physics-based features. By training on these features, our model learns physical boiling laws that statistically describe the manner in which bubbles nucleate, coalesce, and depart under boiling conditions, enabling in situ boiling curve prediction with a mean error of 6%. Our framework offers an automated, learning-based, alternative to conventional boiling heat transfer metrology.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
