Convolutional Neural Network Training Research Articles

Transfer learning classification of suspicious lesions on breast ultrasound: is there room to avoid biopsies of benign lesions?

BackgroundBreast cancer (BC) is the most common malignancy in women and the second cause of cancer death. In recent years, there has been a strong development in artificial intelligence (AI) applications in medical imaging for several tasks. Our aim was to evaluate the potential of transfer learning with convolutional neural networks (CNNs) in discriminating suspicious breast lesions on ultrasound images.MethodsTransfer learning performances of five different CNNs (Inception V3, Xception, Densenet121, VGG 16, and ResNet50) were evaluated on a public and on an institutional dataset (526 and 392 images, respectively), customizing the top layers for the specific task. Institutional images were contoured by an expert radiologist and processed to feed the CNNs for training and testing. Postimaging biopsies were used as a reference standard for classification. The area under the receiver operating curve (AUROC) was used to assess diagnostic performance.ResultsNetworks performed very well on the public dataset (AUROC 0.938–0.996). The direct generalization to the institutional dataset resulted in lower performances (max AUROC 0.676); however, when tested on BI-RADS 3 and BI-RADS 5 only, results were improved (max AUROC 0.792). Good results were achieved on the institutional dataset (AUROC 0.759–0.818) and, when selecting a threshold of 2% for classification, a sensitivity of 0.983 was obtained for three of five CNNs, with the potential to spare biopsy in 15.3%–18.6% of patients.ConclusionIn conclusion, transfer learning with CNNs may achieve high sensitivity and might be used as a support tool in managing suspicious breast lesions on ultrasound images.Relevance statementTransfer learning is a powerful technique to exploit the performances of well-trained CNNs for image classification. In a clinical scenario, it might be useful for the management of suspicious breast lesions on breast ultrasound, potentially sparing biopsy in a non-negligible number of patients.Key PointsProperly trained CNNs with transfer learning are highly effective in differentiating benign and malignant lesions on breast ultrasound.Setting clinical thresholds increased sensitivity.CNNs might be useful as support tools in managing suspicious lesions on breast ultrasound.Graphical

Behavioral Cloning Strategies in Steering Angle Prediction: Applications in Mobile Robotics and Autonomous Driving

Artificial neural networks (ANNs) are artificial intelligence techniques that have made autonomous driving more efficient and accurate; however, autonomous driving faces ongoing challenges in the accuracy of decision making based on the analysis of the vehicle environment. A critical task of ANNs is steering angle prediction, which is essential for safe and effective navigation of mobile robots and autonomous vehicles. In this study, to optimize steering angle prediction, NVIDIA’s architecture was adapted and modified along with the implementation of the Swish activation function to train convolutional neural networks (CNNs) by behavioral cloning. The CNN used human driving data obtained from the UDACITY beta simulator and tests in real scenarios, achieving a significant improvement in the loss function during training, indicating a higher efficiency in replicating human driving behavior. The proposed neural network was validated through implementation on a differential drive mobile robot prototype, by means of a comparative analysis of trajectories in autonomous and manual driving modes. This work not only advances the accuracy of steering angle predictions but also provides valuable information for future research and applications in mobile robotics and autonomous driving. The performance results of the model trained with the proposed CNN show improved accuracy in various operational contexts.

Photodiode-based focus monitoring in ultrashort-pulsed laser structuring of graphite anodes for lithium-ion batteries

In recent years, there has been an increased demand for elaborate monitoring techniques in laser material processing. This has been driven by the need for fast and cost-efficient quality assurance processes. At the same time, ultrashort-pulsed (USP) laser radiation has emerged as a promising technology for creating intricate microstructures in lithium-ion battery graphite anodes due to its high precision and negligible thermal impact. However, the integration of process monitoring in USP laser applications for graphite anode structuring is still unexplored. There is a lack of clarity on suitable sensors, observable parameters, and extractable process-relevant insights. The presented study addressed this gap by demonstrating the capability of state-of-the-art photodiode-based monitoring systems in collecting process-relevant data and deriving valuable insights. A sensor equipped with three photodiodes was employed to address these challenges. Exploratory data analysis and machine learning methodologies were leveraged to develop a data pipeline for processing the acquired information. The data were used to train convolutional neural networks that could accurately predict the focal position. At the same time, the limitations of traditional regression approaches could be shown. The findings advanced the understanding of the possibilities of process monitoring in USP laser applications and emphasized the significance of data-driven approaches in optimizing manufacturing processes.

Reconstruction error based implicit regularization method and its engineering application to lung cancer diagnosis

The automatic diagnosis of lung cancer via artificial intelligence faces two hotspot issues: (1) insufficient data and (2) excessive redundant information, which make it difficult for convolutional neural networks (CNNs) to learn discriminative information of lung cancer. In this paper, we present the reconstruction error based implicit regularization method (REbIRM) that regularizes CNNs at the loss layer. During each training iteration, the reconstruction errors introduced by the two-stage discriminative auto-encoder are used to sharpen the generalization ability of deep CNNs by improving the decision boundary. In the application process, the trained deep CNN is used for completing computed tomography (CT) diagnostics. The main clinical benefit of our approach is that it is domain independent, requiring no specialized knowledge, and can therefore be applied to different types of datasets. To the best of our knowledge, this is the first attempt to implicitly regularize CNNs based on the reconstruction errors. Finally, experimental results on three CT image classification datasets show that REbIRM can achieve impressive results and that, in conjunction with Dropout, it obtains the state-of-the-art performance. REbIRM is also robust to the selection of hyper-parameters and only has the sublinear influence on the convergence of deep CNNs. Besides, empirical and theoretical evidence are provided to indicate that REbIRM prefers to converges in a constrained parameter space with flatter minima, which explains why it can generalize to new data. Finally, the nature of REbIRM is further explored through visualization techniques to analyze how it works in training deep CNNs.

PV Module Soiling Detection Using Visible Spectrum Imaging and Machine Learning

During the last decades photovoltaic solar energy has continuously increased its share in the electricity mix and has already surpassed 5% globally. Even though photovoltaic (PV) installations are considered to require very little maintenance, their efficient exploitation relies on accounting for certain environmental factors that affect energy generation. One of these factors is the soiling of the PV surface, which could be observed in different forms, such as dust and bird droppings. In this study, visible spectrum data and machine learning algorithms were used for the identification of soiling. A methodology for preprocessing the images is proposed, which puts focus on any soiling of the PV surface. The performance of six classification machine learning algorithms is evaluated and compared—convolutional neural network (CNN), support vector machine (SVM), random forest (RF), k-nearest neighbor (kNN), naïve-Bayes, and decision tree. During the training and validation phase, RF proved to be the best-performing model with an F1 score of 0.935, closely followed by SVM, CNN, and kNN. However, during the testing phase, the trained CNN achieved the highest performance, reaching F1 = 0.913. SVM closely followed it with a score of 0.895, while the other two models returned worse results. Some results from the application of the optimal model after specific weather events are also presented in this study. They confirmed once again that the trained convolutional neural network can be successfully used to evaluate the soiling state of photovoltaic surfaces.

Concrete crack classification based on fourier image enhancement and convolutional neural network

This paper investigates the application of Fourier image enhancement combined with Convolutional Neural Networks (CNNs) for detecting cracks in concrete structures. Fourier enhancement is used to preprocess crack images, improving their clarity and reducing noise, which in turn enhances the performance of the CNN in accurately classifying cracks. The results demonstrate that this combination improves the classification accuracy, with the enhanced images achieving a higher accuracy compared to non-enhanced images. Additionally, the study examines the effects of this preprocessing on CNN training time. However, accuracy varies depending on the dataset used, with one dataset reaching a maximum accuracy of 95% after enhancement. These findings highlight the potential of using frequency-domain image enhancement techniques in conjunction with deep learning models for structural health monitoring.

Characterization of AI-enhanced magnetophoretic transistors operating in a tri-axial magnetic field for on-chip bioparticle sorting.

We demonstrate two general classes of magnetophoretic transistors, called the "trap" and the "repel-and-collect" transistors, capable of switching single magnetically labeled cells and magnetic particles between different paths in a microfluidic chamber. Compared with prior work on magnetophoretic transistors operating in a two-dimensional in-plane rotating field, the use of a tri-axial magnetic field has the fundamental advantages of preventing particle cluster formation and better syncing of single particles with the general operating clock. We use finite element methods to investigate the energy distribution on the chip surface and to predict the particle behavior at various device geometries. We then fabricate the proposed transistors and compare the experimental results with the simulation predictions. We found that with gate electrical currents of ~ 40mA for a transistor with proper geometry, complete switching of magnetic particles with diameters in the range of 8-15μm is achieved. We show that the device is reliable and works well at different magnetic field strengths (50-100 Oe) and frequencies (0.05-0.5Hz). We also employed an image processing code with a trained convolutional neural network to automate the proposed transistors for identifying and sorting particles with various sizes and magnetic susceptibilities with accuracies higher than 98%. The proposed transistors can be used in designing novel magnetophoretic circuits for important applications in biomedical microdevices and single-cell biology.

Low-complexity short-time acoustic scene classification based on causal convolutions

Acoustic scene classification is vital for building smart cities. Most acoustic scene classification studies are performed by training convolutional neural networks. The networks constructed based on the general convolutional approach need to consider the temporal-frequency connections in the audio spectrogram. This work proposes a bottleneck structure neural network model for acoustic scene classification based on causal convolution and Sub-spectral normalization to overcome the abovementioned problem. To obtain more sample data, a new mix-spec augmentation data enhancement method is proposed by drawing on the existing classical mixup and spec augmentation methods, and the three are combined for sample augmentation. It was experimentally found that the proposed bottleneck structure helps improve the recognition accuracy and robustness of the model. In addition, the ablation experiment showed that the data augmentation training model combined with the mixup, SpecAugment, and the newly proposed mix-SpecAugment method had the highest classification accuracy. Finally, the near-square audio spectrogram has better training results by employing different shapes of audio spectrograms for the experiments. Compared with the baseline results, it is found that our proposed model significantly improved the accuracy of the baseline and reduced loss.

Privacy-preserving and verifiable convolution neural network inference and training in cloud computing

With the rapid development of cloud computing, outsourcing massive data and complex deep learning model to cloud servers (CSs) has become a popular trend, which also brings some security problems. One is that the model stored in the CSs may be corrupted, leading to incorrect inference and training results. The other is that the privacy of outsourced data and model may be compromised. However, existing privacy-preserving and verifiable inference schemes suffer from low detection probability, high communication overhead and substantial computational time. To solve the above problems, we propose a privacy-preserving and verifiable scheme for convolutional neural network inference and training in cloud computing. In our scheme, the model owner generates the authenticators for model parameters before uploading the model to CSs. In the phase of model integrity verification, model owner and user can utilize these authenticators to check model integrity with high detection probability. Furthermore, we design a set of privacy-preserving protocols based on replicated secret sharing for both the inference and training phases, significantly reducing communication overhead and computational time. Through security analysis, we demonstrate that our scheme is secure. Experimental evaluations show that the proposed scheme outperforms existing schemes in privacy-preserving inference and model integrity verification.

Predicting standardized uptake value of brown adipose tissue from CT scans using convolutional neural networks.

The standard method for identifying active Brown Adipose Tissue (BAT) is [18F]-Fluorodeoxyglucose ([18F]-FDG) PET/CT imaging, which is costly and exposes patients to radiation, making it impractical for population studies. These issues can be addressed with computational methods that predict [18F]-FDG uptake by BAT from CT; earlier population studies pave the way for developing such methods by showing some correlation between the Hounsfield Unit (HU) of BAT in CT and the corresponding [18F]-FDG uptake in PET. In this study, we propose training convolutional neural networks (CNNs) to predict [18F]-FDG uptake by BAT from unenhanced CT scans in the restricted regions that are likely to contain BAT. Using the Attention U-Net architecture, we perform experiments on datasets from four different cohorts, the largest study to date. We segment BAT regions using predicted [18F]-FDG uptake values, achieving 23% to 40% better accuracy than conventional CT thresholding. Additionally, BAT volumes computed from the segmentations distinguish the subjects with and without active BAT with an AUC of 0.8, compared to 0.6 for CT thresholding. These findings suggest CNNs can facilitate large-scale imaging studies more efficiently and cost-effectively using only CT.

CLASSIFICATION OF ACUTE LYMPHOBLASTIC LEUKEMIA CELLS USING ARTIFICIAL INTELLIGENCE

Due to the morphological similarity between immature lymphoblasts (cancerous cells) to lymphocytes (non-cancerous cells), detecting Acute Lymphoblastic Leukemia poses a significant challenge for pathologists. These cells, which exhibit a similar pattern, can lead to various errors during the diagnosis of the disease. In this study, the cancerous and non-cancerous cells were classified using 3 different artificial intelligence approaches. In the first approach, the classification process was carried out by training Convolutional Neural Networks in 4 different architectures. In the second approach, a hybrid approach was proposed by combining the convolution layer of the CNN model as the feature extractor with the Support Vector Machine, Naive Bayes and Random Forest algorithms as the classifier. The classification processes were carried out by training the proposed second approach. In the third approach, the classification process was performed using transfer learning process and ResNet50 and VGG16 networks. In all experiments, the effects of hyper-parameter and dataset changes on model performance were also examined. The results obtained by these three approaches were compared using the Accuracy, Precision, Recall, F-score, and AUC performance measures. It was determined that the most successful results were obtained with the 1st approach using the Dataset3.

Retrieval of boundary layer precipitable water from GOES ABI using machine learning techniques

Abstract Low-level moisture is an important ingredient for forecasting severe storms, especially over the Great Plains where severe storms often develop along the dryline. Although ground-based observation systems such as lidar or radiosondes on weather balloons provide accurate information on low-level moisture, data are provided at limited locations, and the low temporal resolution of radiosondes makes it difficult to track a rapidly developing dryline. Geostationary satellites provide high spatial and temporal observation, but the channels of current geostationary satellites are mostly sensitive to water vapor at mid to upper levels. However, the split window difference (SWD) between the “clean” window channel (10.3 μm) and the “dirty” window channel (12.3 μm) is commonly used for estimating low-level water vapor. However, this estimation is complicated by surface temperature contributions, dependence on the lapse rate, and nonlinear relationships between SWD and moisture. This study applies machine learning techniques to infer boundary layer precipitable water (BLPW) from Geostationary Operational Environmental Satellite (GOES) Advanced Baseline Imager (ABI) data. Since there are few observations that cover wide regions for training convolutional neural networks, especially for the atmosphere above the surface, High-Resolution Rapid Refresh (HRRR) model outputs are used as the truth for training. Statistical results show that using ABI channel 13, 14, and 15 brightness temperatures, skin temperature, solar zenith angle, and GOES total precipitable water product as additional inputs show the lowest MSE. Case study results show that the model developed in this study is good at capturing strong moisture gradients and depicting a dryline.

Deep learning from three-dimensional lithium-ion battery multiphysics model part I: Data development

Fast growing demands for electric vehicles require better longevity, safety and reliability for next-generation high-energy battery technologies. A data-centered battery management system is thus desired to interpret complex battery data and make decisions for properly managing multi-physics battery dynamics. Nowadays, Battery informatics are emerging as promising solutions by leveraging advanced machine learning tools to deliver accurate prediction of battery performance, health and safety, but is hurdled by a scarcity of data. To mitigate this issue, this study presents one of the first studies for data development through both experimental studies and three-dimensional (3-D) multi-physics modeling to underpin a deep learning framework with in-depth examination for battery performance and thermal risk prediction. Specifically, Part I focused on the development of the battery model which was thoroughly validated and analyzed to guarantee the model accuracy by two steps: firstly, we validated the multi-physics model against two commercial Lithium-ion batteries, i.e., Panasonic NCR18650B and 18650BD; Then, the coupling between thermal and electrochemical battery behaviors were analyzed deeply to demonstrate insights obtained from the model, such as voltage evolution and maximum local temperature (hot spot). The developed model proves to be capable of providing insightful and reliable data for the training of convolutional neural network and long short-term memory (CNN-LSTM) in part II.

Physics-informed convolutional neural network for microgrid economic dispatch

The variability of renewable energy generation and the unpredictability of electricity demand create a need for real-time economic dispatch (ED) of assets in microgrids. However, solving numerical optimization problems in real-time can be incredibly challenging. This study proposes using a convolutional neural network (CNN) based on deep learning to address these challenges. Compared to traditional methods, CNN is more efficient, delivers more dependable results, and has a shorter response time when dealing with uncertainties. While CNN has shown promising results, it does not extract explainable knowledge from the data. To address this limitation, a physics-inspired CNN model is developed by incorporating constraints of the ED problem into the CNN training to ensure that the model follows physical laws while fitting the data. The proposed method can significantly accelerate real-time economic dispatch of microgrids without compromising the accuracy of numerical optimization techniques. The effectiveness of the proposed data-driven approach for optimal allocation of microgrid resources in real-time is verified through a comprehensive comparison with conventional numerical optimization approaches.

Deep learning-driven optimization of real-time ray tracing for enhanced immersive virtual reality

Photorealistic rendering is essential for immersive virtual reality and real-time ray tracing is required for this, yet the necessary computational load has so far hindered its use in scenarios where a fast response and high frame rates are paramount. To overcome this challenge, we present a deep learning-based model that optimises the computational load of ray tracing through utilising convolutional neural networks (CNNs) to predict lightscene interactions. We achieve this by training CNNs from datasets with offline ray-traced images. Using this learning process, we approximate the contents of light transport simulations from scene point sampling. The use of this optimised algorithm in virtual reality scenes, thus, enables renderings of more complex scenes, with dynamic lighting, complex geometry and interactive elements under high frame rates, keeping users immersed and responsive to events. Our results show that this method achieves visual similarities to traditional ray tracing, which is photorealistic rendering, but can be performed in real-time in virtual reality. We anticipate this work leads to many potential applications of ray tracing in many scenarios in virtual reality, such as gaming, architectural rendering and training.

Vibration based dual-criteria damage detection method using deep neural networks in highway bridges with steel girders

This article introduces a new method for assessing the damage in steel girders of highway bridges using deep neural networks. The method uses modified forms of modal strain energy damage index and modal flexibility damage index to train the neural network and prevent unsafe decisions. The neural network is trained using damage indexes achieved from numerical simulations of the bridge with different damage scenarios. Three types of neural networks, including convolutional neural network (CNN), Long short-term memory (LSTM) neural network, and conventional artificial neural network (ANN) are compared to achieve the best performance. The proposed method overcomes previous detection problems such as false-positive indications, limited damage locations, and scenarios, and insufficient accuracy in cases with multiple damage scenarios. The results show that the proposed method and CNN can accurately identify unspecified damage locations and severity in multi-span highway bridges. The new training method of the CNN deep neural network systems overcomes some shortcomings in ANN, such as being time-consuming, having a low convergence rate, and experiencing data overfitting. Additionally, the CNN training scheme can limit the need for massive amounts of input data and increase the speed and accuracy of network training, especially in multiple damage scenarios.

DropletMask: Leveraging visual data for droplet impact analysis

AbstractMachine learning‐assisted computer vision represents a state‐of‐the‐art technique for extracting meaningful features from visual data autonomously. This approach facilitates the quantitative analysis of images, enabling object detection and tracking. In this study, we utilize advanced computer vision to precisely identify droplet motions and quantify their impact forces with spatiotemporal resolution at the picoliter or millisecond scale. Droplets, captured by a high‐speed camera, are denoised through neuromorphic image processing. These processed images are employed to train convolutional neural networks, allowing the creation of segmented masks and bounding boxes around moving droplets. The trained networks further digitize time‐varying multi‐dimensional droplet features, such as droplet diameters, spreading and sliding motions, and corresponding impact forces. Our innovative method offers accurate measurement of small impact forces with a resolution of approximately 10 pico‐newtons for droplets in the micrometer range across various configurations with the time resolution at hundreds of microseconds.

A Study on Resolving Dynamic analysis using deep learning Framework Logical Methods

Deep learning technologies have gained a great deal of interest in recent months for demonstrating material removal capabilities in a variety of separation, categorization, as well as other machine vision activities. The majority of these deep networks are built employing convolution or completely convolution structures. Our suggested Dynamic and Static gestures System guided by a hand gesture high dimensional data methodology We use a snapshot of a human's thoracic spine to determine the person's depth and the size of the space around his or her hands. In this research, we provide a novel object-based deep-learning framework for semantic segmentation of extremely high-resolution satellite data. In specifically, we leverage object-based prior record by augmenting a fully convolutional neural network's training strategy with an anisotropic diffusion data pre-processing phase and an additional loss term. The goal of this restricted framework is to ensure that similar visual data is assigned to the same semantic class. Here, we employ intermediate steps based on traditional image processing techniques to aid in the resolution of the subsequent problem of classification, detection, or segmentation. Research shows that adding pre- and post-processing steps to a deep learning pipeline can boost model performance over using only the network. Recent advances in UQ algorithms used for deep learning are analysed, and their potential for use in relevance feedback is addressed. It also emphasizes basic research difficulties and directions linked with UQ. Quantifiably, man-made classes with more precise geometry, such as buildings, benefited the most from our strategy, particularly along object borders, indicating the developed approach's enormous potential. The purpose of this paper is to offer an overview of the approaches used inside deep learning frameworks to either appropriately prepare the input (pre-processing) or enhance the outcomes of the network output (post-processing), with an emphasis on digital pathology image analysis.

Recognizing Digital Ink Chinese Characters Written by International Students Using a Residual Network with 1-Dimensional Dilated Convolution

Due to the complex nature of Chinese characters, junior international students often encounter writing problems related to strokes, components, and their combinations when writing Chinese characters. Digital ink Chinese characters (DICCs) are obtained by sampling the writing trajectory of Chinese characters with a pen input device. DICCs contain rich information, such as the time and space of strokes and sampling points. Recognizing DICCs is crucial for evaluating and correcting writing errors and enhancing the quality of Chinese character teaching for international students. Here, the paper first employs a one-dimensional dilated convolution to digital ink Chinese character recognition (DICCR) and proposes a novel residual network with one-dimensional dilated convolution (1-D ResNetDC). The 1-D ResNetDC not only utilizes multi-scale convolution kernels, but also employs different dilation rates on a single-scale convolution kernel to obtain information from various ranges. Additionally, residual connections facilitate the training of deep one-dimensional convolutional neural networks. Moreover, the paper proposes a more expressive ten-dimensional feature representation that includes spatial, temporal, and writing direction information for each sampling point, thereby improving classification accuracy. Because the DICC dataset of international students is small and unbalanced, the 1-D ResNetDC is pre-trained on the published available dataset. The experiments demonstrate that our approach is effective and superior. This model features a compact architecture, a reduced number of parameters, and excellent scalability.

Real-Time CNN Training and Compression for Neural-Enhanced Adaptive Live Streaming.

We propose a real-time convolutional neural network (CNN) training and compression method for delivering high-quality live video even in a poor network environment. The server delivers a low-resolution video segment along with the corresponding CNN for super resolution (SR), after which the client applies the CNN to the segment in order to recover high-resolution video frames. To generate a trained CNN corresponding to a video segment in real-time, our method rapidly increases the training accuracy by promoting the overfitting property of the CNN while also using curriculum-based training. In addition, assuming that the pretrained CNN is already downloaded on the client side, we transfer only residual values between the updated and pretrained CNN parameters. These values can be quantized with low bits in real time while minimizing the amount of loss, as the distribution range is significantly narrower than that of the updated CNN. Quantitatively, our neural-enhanced adaptive live streaming pipeline (NEALS) achieves higher SR accuracy and a lower CNN compression loss rate within a constrained training time compared to the state-of-the-art CNN training and compression method. NEALS achieves 15 to 48% higher quality of the user experience compared to state-of-the-art neural-enhanced live streaming systems.