Lung cancer ranks among the most common types of cancer. Noninvasive computer-aided diagnosis can enable large-scale rapid screening of potential patients with lung cancer. Deep learning methods have already been applied for the automatic diagnosis of lung cancer in the past. Due to restrictions caused by single modality images of dataset as well as the lack of approaches that allow for a reliable extraction of fine-grained features from different imaging modalities, research regarding the automated diagnosis of lung cancer based on noninvasive clinical images requires further study. In this paper, we present a deep learning architecture that combines the fine-grained feature from PET and CT images that allow for the noninvasive diagnosis of lung cancer. The multidimensional (regarding the channel as well as spatial dimensions) attention mechanism is used to effectively reduce feature noise when extracting fine-grained features from each imaging modality. We conduct a comparative analysis of the two aspects of feature fusion and attention mechanism through quantitative evaluation metrics and the visualization of deep learning process. In our experiments, we obtained an area under the ROC curve of 0.92 (balanced accuracy = 0.72) and a more focused network attention which shows the effective extraction of the fine-grained feature from each imaging modality.

Printed circuit board wire segmentation based on computed tomography (CT) image can help subsequently locate and estimate inner faults of circuit in an automatic and non‐destructive manner. However, CT imaging is prone to suffer from scattered artefacts, metal artefacts and other interference, destroying compact boundary structures of wires. Wires have the characteristic of dense local distribution, and massive vias, pads, and coppers can appear close to wires, resulting in mazy recognition surroundings. The above‐mentioned problems bring great difficulty for high‐accuracy recognition and location of wire segmentation. In this study, considering that deep convolutional neural network (DCNN) with powerful feature representation can recognise wires in confused surroundings, and graph cut (GC) model relying on grayscale and local texture information specialises in protecting edge structures of wires, the authors propose an effective framework called DCNN‐GC that employs DCNN to obtain global semantic prior to guide the GC model to accomplish satisfactory wire segmentation. The authors qualitative and quantitative results demonstrate outstanding performance, and achieve overwhelming intersection over union compared with traditional and DCNN‐based methods.

The mutation status of the epidermal growth factor receptor (EGFR) is an important clinical reference indicator for lung cancer diagnosis and treatment. However, the extraction of effective discriminative features for non-invasive computer-aided EGFR mutation prediction still poses a big challenge. In this paper, multiple types of features are designed and analyzed to address this problem. These features include clinical features based on prior medical knowledge and quantitative image features extracted by convolutional neural networks (CNN). A long short-term memory (LSTM) network is also introduced to exploit the dependency between these feature types and then fuse them. In particular, a CNN is constructed to extract quantitative features of computed-tomography (CT) images. Furthermore, a LSTM is utilized to analyze the dependency between these clinical and CT image features and generate a new feature representation for computer-aided diagnosis. For samples from the same category, the proposed method deal with feature representation variabilities arising from interdependencies in multi-type features and patient specificity. The multiple feature types of the collected clinical data are used to assess the proposed approach and other relevant algorithms. Our results demonstrate that the multi-type dependency-based feature representation shows superior performance (Accuracy = 75%, AUC = 0.78) compared to single-type feature representations. The proposed method is reliable to apply for diagnosing of the EGFR mutation status.

In image classification of deep learning, adversarial examples where input is intended to add small magnitude perturbations may mislead deep neural networks (DNNs) to incorrect results, which means DNNs are vulnerable to them. Different attack and defense strategies have been proposed to better research the mechanism of deep learning. However, those researches in these networks are only for one aspect, either an attack or a defense. There is in the improvement of offensive and defensive performance, and it is difficult to promote each other in the same framework. In this paper, we propose Cycle-Consistent Adversarial GAN (CycleAdvGAN) to generate adversarial examples, which can learn and approximate the distribution of the original instances and adversarial examples, especially promoting attackers and defenders to confront each other and improve their ability. For CycleAdvGAN, once the GeneratorA and D are trained, GA can generate adversarial perturbations efficiently for any instance, improving the performance of the existing attack methods, and GD can generate recovery adversarial examples to clean instances, defending against existing attack methods. We apply CycleAdvGAN under semiwhite-box and black-box settings on two public datasets MNIST and CIFAR10. Using the extensive experiments, we show that our method has achieved the state-of-the-art adversarial attack method and also has efficiently improved the defense ability, which made the integration of adversarial attack and defense come true. In addition, it has improved the attack effect only trained on the adversarial dataset generated by any kind of adversarial attack.