Convolutional neural networks (CNNs) have made remarkable achievements in many tasks. However, most of them are hardly applied to embedded systems directly because of the requirement of huge memory ...space and computing power. In this article, we propose a pruning framework, namely, FiltDivNet, to accelerate and compress CNN models for their applicability to small or portable devices. The correlations among filters are taken into account and measured by the goodness of fit. On this basis, a hybrid-cluster pruning strategy is designed with dynamic pruning ratios for different clusters in CNN models. It aims at representing its filters in their diversity by removing redundant ones cluster by cluster. In addition, a new loss function with adaptive sparsity constraints is introduced for the retraining and fine-tuning in the FiltDivNet. Finally, some comparative experiments based on classical CNN models are carried out to demonstrate its effectiveness in compression performance and its adaptability with different CNN architectures.
•Convolutional neural networks can efficiently segment breast masses in ultrasound.•Segmentation network's receptive field can be adjusted with an attention mechanism.•Segmentation performance ...assessment based on multiple datasets.
In this work, we propose a deep learning method for breast mass segmentation in ultrasound (US). Variations in breast mass size and image characteristics make the automatic segmentation difficult. To address this issue, we developed a selective kernel (SK) U-Net convolutional neural network. The aim of the SKs was to adjust network's receptive fields via an attention mechanism, and fuse feature maps extracted with dilated and conventional convolutions. The proposed method was developed and evaluated using US images collected from 882 breast masses. Moreover, we used three datasets of US images collected at different medical centers for testing (893 US images). On our test set of 150 US images, the SK-U-Net achieved mean Dice score of 0.826, and outperformed regular U-Net, Dice score of 0.778. When evaluated on three separate datasets, the proposed method yielded mean Dice scores ranging from 0.646 to 0.780. Additional fine-tuning of our better-performing model with data collected at different centers improved mean Dice scores by ∼6%. SK-U-Net utilized both dilated and regular convolutions to process US images. We found strong correlation, Spearman's rank coefficient of 0.7, between the utilization of dilated convolutions and breast mass size in the case of network's expansion path. Our study shows the usefulness of deep learning methods for breast mass segmentation. SK-U-Net implementation and pre-trained weights can be found at github.com/mbyr/bus_seg.
The establishment of image correspondence through robust image registration is critical to many clinical tasks such as image fusion, organ atlas creation, and tumor growth monitoring and is a very ...challenging problem. Since the beginning of the recent deep learning renaissance, the medical imaging research community has developed deep learning-based approaches and achieved the state-of-the-art in many applications, including image registration. The rapid adoption of deep learning for image registration applications over the past few years necessitates a comprehensive summary and outlook, which is the main scope of this survey. This requires placing a focus on the different research areas as well as highlighting challenges that practitioners face. This survey, therefore, outlines the evolution of deep learning-based medical image registration in the context of both research challenges and relevant innovations in the past few years. Further, this survey highlights future research directions to show how this field may be possibly moved forward to the next level.
In the last five years, deep learning has been introduced to tackle the hyperspectral image (HSI) classification and demonstrated good performance. In particular, the convolutional neural network ...(CNN)-based methods for HSI classification have made great progress. However, due to the high dimensionality of HSI and equal treatment of all bands, the performance of these methods is hampered by learning features from useless bands for classification. Moreover, for patchwise-based CNN models, equal treatment of spatial information from the pixel-centered neighborhood also hinders the performance of these methods. In this article, we propose an end-to-end residual spectral-spatial attention network (RSSAN) for HSI classification. The RSSAN takes raw 3-D cubes as input data without additional feature engineering. First, a spectral attention module is designed for spectral band selection from raw input data by emphasizing useful bands for classification and suppressing useless bands. Then, a spatial attention module is designed for the adaptive selection of spatial information by emphasizing pixels from the same class as the center pixel or those are useful for classification in the pixel-centered neighborhood and suppressing those from a different class or useless. Second, two attention modules are also used in the following CNN for adaptive feature refinement in spectral-spatial feature learning. Third, a sequential spectral-spatial attention module is embedded into a residual block to avoid overfitting and accelerate the training of the proposed model. Experimental studies demonstrate that the RSSAN achieved superior classification accuracy compared with the state of the art on three HSI data sets: Indian Pines (IN), University of Pavia (UP), and Kennedy Space Center (KSC).
Although automated Skin Lesion Classification (SLC) is a crucial integral step in computer-aided diagnosis, it remains challenging due to variability in textures, colors, indistinguishable ...boundaries, and shapes.
This article proposes an automated dermoscopic SLC framework named Dermoscopic Expert (DermoExpert). It combines the pre-processing and hybrid Convolutional Neural Network (hybrid-CNN). The proposed hybrid-CNN has three distinct feature extractor modules, which are fused to achieve better-depth feature maps of the lesion. Those single and fused feature maps are classified using different fully connected layers, then ensembled to predict a lesion class. In the proposed pre-processing, we apply lesion segmentation, augmentation (geometry- and intensity-based), and class rebalancing (penalizing the majority class’s loss and merging additional images to the minority classes). Moreover, we leverage transfer learning from the pre-trained models. Finally, we deploy the weights of our DermoExpert to a possible web application.
We evaluate our DermoExpert on the ISIC-2016, ISIC-2017, and ISIC-2018 datasets, where the DermoExpert has achieved the area under the receiver operating characteristic curve (AUC) of 0.96, 0.95, and 0.97, respectively. The experimental results improve the state-of-the-art by the margins of 10.0% and 2.0%, respectively, for the ISIC-2016 and ISIC-2017 datasets in terms of AUC. The DermoExpert also outperforms by 3.0% for the ISIC-2018 dataset concerning a balanced accuracy.
Since DermoExpert provides better classification outcomes on three different datasets, leading to a better recognition tool to assist dermatologists. Our source code and segmented masks for the ISIC-2018 dataset will be available as a public benchmark for future improvements.
Display omitted
•Proposing a hybrid-CNN classifier for multiple skin diseases recognition.•Precisely segmenting skin lesion although the presence of hair fibers and other artifacts.•Class-rebalancing, transfer learning, and augmentation for a generic model, as tiny datasets are being used.•State of the art results on ISIC-16 (2-class), ISIC-17 (3-class), and ISIC-18 (7-class).•Development of a possible web application, deploying our trained model’s weights.
Hyperspectral image change detection (HSI-CD) is a fundamental task in the field of remote sensing (RS) observation, which utilizes the rich spectral and spatial information in bitemporal HSIs to ...detect subtle changes on the Earth's surface. However, modern deep learning (DL)-based HSI-CD methods mostly rely on patch-based methods, which leads to spectral band redundancy and spatial information noise in limited receiving domains, thus ignoring the extraction and utilization of saliency information and limiting the improvement of CD performance. To address these issues, this article proposes a joint saliency temporal-spatial-spectral information network (STSS-Net) for HSI-CD. The principal contributions of this article can be summarized: 1) we have designed a spatial saliency information extraction (SSIE) module for denoising based on distance from center pixels and spectral similarity of the substance, which increases the attention to spatial differences between similar spectral substances and different spectral substances; 2) we have designed a compact high-level spectral information tokenizer (CHLSIT) for spectral saliency information, where the high-level conceptual information of changes in spectral interest can be represented by nonlinear combinations of spectral bands, and redundancy can be removed by extracting high-level spectral conceptual features; and 3) utilizing the advantages of CNN and transformer architectures to combine temporal-spatial-spectral information. The experimental results on three real HSI-CD datasets show that STSS-Net can improve the accuracy of CD and has a certain improvement in the detection of edge information and complex information.
The use of convolutional neural networks (CNNs) in seismic interpretation tasks, like facies classification, has garnered a lot of attention for its high accuracy. However, its drawback is usually ...poor generalization when trained with limited training data pairs, especially for noisy data. Seismic images are dominated by diverse wavelet textures corresponding to seismic facies with various petrophysical parameters, which can be suitably represented by Gabor functions. Inspired by this fact, we propose using learnable Gabor convolutional kernels in the first layer of a CNN network to improve its generalization. The modified network combines the interpretability features of Gabor filters and the reliable learning ability of the original CNN. It replaces the pixel nature of conventional CNN filters with a constrained function form that depends on five parameters that are more in line with seismic signatures. This allows us, in training, to constrain the angle and wavelength of the Gabor kernels to specific ranges to help enhance the seismic features and reduce noise. We, also, test this modified CNN using various kernels on salt&pepper and speckle noise. The experiments on the Netherland F3 dataset show that we obtain the best generalization and robustness of the CNN to noise when Gabor kernels are used in the first layer.
•The proposed denoising method can eliminate severe noise in a wrapped phase. In the current network, noise level up to a signal to noise ratio (SNR) value of −4 dB can be successfully removed.•After ...CNN denoising, a wrapped phase can be easily unwrapped using an existing simple unwrapping method.•The proposed denoising method is shown to be effective through extensive comparisons with the average filter, median filter and windowed Fourier filtering. The proposed method outperforms the conventional average filter and median filter, and performs similarly to windowed Fourier filtering. The feasibility of phase unwrapping strategy is demonstrated by comparing with an existing deep learning unwrapping method.•The proposed denoising method is able to directly provide denoising result without adjusting any parameters.
We propose a wrapped phase denoising method based on convolutional neural networks (CNN), which can effectively denoise a noisy wrapped phase. The noisy numerator and denominator of the arctangent function are firstly denoised by CNN, and then the filtered numerator and denominator use the arctangent function to obtain the clean wrapped phase. We experimentally verify the denoising performance using various wrapped phase that contains different noise conditions, where the denoised wrapped phase can achieve a satisfactory unwrapping performance using the existing simple unwrapping method. In addition, the proposed method is further demonstrated through the comparison of the existing methods, and shows an accurate denoising result without adjusting any parameters.
Previous 1-D convolutional neural network (1-D CNN) models for vibration fault diagnosis have high computational complexity and poor interpretability, which cannot meet the higher requirements of ...model storage, computational efficiency, and reliability for airborne and portable devices. Considering these challenges, an explainable and lightweight 1-D CNN (ELCNN) model based on square global average pooling (S-GAP) and improved vibration signals is proposed. The feature extraction and classification layers of 1-D CNN are optimized to minimize the model parameters and computational complexity and improve interpretability while ensuring diagnostic accuracy. The model compresses the number of convolutional layers, removes unnecessary bias, activation function, and pooling layer, and replaces the fully connected layer (FCL) with S-GAP. Improved 1-D CNN models of different methods are evaluated and analyzed on public datasets of rolling bearings. Results show that the ELCNN improved for vibration signals is more lightweight, antinoise, and explainable than other models, and the diagnostic accuracy is further improved.