The application of deep convolutional neural networks (CNN) in the field of medical image processing has attracted extensive attention and demonstrated remarkable progress. An increasing number of deep learning methods have been devoted to classifying ChestX-ray (CXR) images, and most of the existing deep learning methods are based on classic pretrained models, trained by global ChestX-ray images. In this paper, we are interested in diagnosing ChestX-ray images using our proposed Fusion High-Resolution Network (FHRNet). The FHRNet concatenates the global average pooling layers of the global and local feature extractors—it consists of three branch convolutional neural networks and is fine-tuned for thorax disease classification. Compared with the results of other available methods, our experimental results showed that the proposed model yields a better disease classification performance for the ChestX-ray 14 dataset, according to the receiver operating characteristic curve and area-under-the-curve score. An ablation study further confirmed the effectiveness of the global and local branch networks in improving the classification accuracy of thorax diseases.

Computed Tomography (CT) imaging is one of the most widely-used and cost-effective technology for organ screening and diseases diagnosis. Because of existence of metallic implants in some patients, the CT images acquired from these patients are often corrupted by undesirable metal artifacts, which causes severe problem of metal artifact. Although there have been proposed many methods to reduce metal artifact, reduction is still challenging and inadequate, and results are suffering from symptom variance, second artifact and poor subjective evaluation. To address these problems, we propose a novel metal artifact reduction method based on generative adversarial networks to simultaneously reduce metal artifacts and enhance texture structure of corrected CT images. Specifically, we firstly incorporate interactive information (text) and imaging CT (image) into a comprehensive feature to yield multi-modal feature-fusion representation, which overcomes the representative ability limitation of single-modal data. The incorporation of interaction information constrains the feature generation to ensure symptom consistency between corrected and target CT. Then, we design an edge-enhance sub-network to avoid second artifact and suppress noise. Besides, we invite three professional physicians to evaluate corrected CT image subjectively. In this paper, We achieved average increment of 11.3% PSNR and 12.1% SSIM on DeepLesion dataset. The subjective evaluations by physicians show that ours outperforms over 6.3%, 7.1%, 5.50% and 6.9% in term of sharpness, resolution, invariance and acceptability, respectively. Our proposed method can achieve high-quality metal artifact reduction results.

The existing medical imaging technologies have little consideration on color information, thus most of medical images are gray. Classical hand-craft features-based methods have obtained unsatisfactory results in colorizing medical images. Moreover, these methods ignore the deep feature of medical images that represent pathology and color information. In this paper, we propose a novel method that iteratively colorizes grayscale medical images under preserving content in fine-tuned deep neural network. To the best of our knowledge, there is no currently work that attempts to colorize the medical image by using deep neural network. Specifically, we propose Y-loss which is defined as nonlinear combination of ℓ1 and ℓ2 norm to preserve content invariance between target and colorized medical image. Then, adaptive reference image search algorithm is introduced to code reference and target medical image with D-hash and search reference image in hash code automatically, which free the manual selection of the reference image. Extensive experiment results show that the proposed method can generate higher quality colored medical image than recent state-of-the-art methods, and can be approved by the doctor. The objective evaluation (PSNR and SSIM) outperform an average increment 24% and 47% than baseline method, respectively. Our code is available at: https://github.com/Tongshiyue/Adaptive-medical-image-deep-color-perception-algorithm.

Background: Cardiovascular disease not only occurs in the elderly but also tends to become a common social health problem. Considering the fast pace of modern life, quantified heart rate variability (HRV) indicators combined with the convenience of wearable devices are of great significance for intelligent telemedicine. To quantify the changes in human mental state, this article proposes an improved differential threshold algorithm for R-wave detection and recognition of electrocardiogram (ECG) signals.Methods: HRV is a specific quantitative indicator of autonomic nerve regulation of the heart. The recognition rate is increased by improving the starting position of R wave and the time-window function of the traditional differential threshold method. The experimental platform is a wearable sign monitoring system constructed based on body area networks (BAN) technology. Analytic hierarchy process (AHP) is used to construct the mental stress assessment model, the weight judgment matrix is constructed according to the influence degree of HRV analysis parameters on mental stress, and the consistency check is carried out to obtain the weight value of the corresponding HRV analysis parameters.Results: Experimental results show that the recognition rate of R wave of real-time 5 min ECG data collected by this algorithm is >99%. The comprehensive index of HRV based on weight matrix can greatly reduce the deviation caused by the measurement error of each parameter. Compared with traditional characteristic wave recognition algorithms, the proposed algorithm simplifies the process, has high real-time performance, and is suitable for wearable analysis devices with low-configuration requirements.Conclusion: Our algorithm can describe the mental stress of the body quantitatively and meet the requirements of application demonstration.

Objective. Spread through air space (STAS) is an invasive characterization of lung adenocarcinoma and is regarded as a risk factor for poor prognosis. The aim of this study is to develop a random forest model for preoperative prediction of spread through air spaces (STAS) in stage IA lung adenocarcinoma. Methods. 92 patients with stage IA lung adenocarcinoma, who underwent computed tomography (CT) scan and surgical resection, were retrospectively reviewed. Each pulmonary nodule was automatically segmented by artificial intelligence (AI) software, and its CT-based radiomics were extracted. All patients were pathologically classified into STAS-negative and STAS-positive cohorts; then, clinical pathological and CT-based radiomics were compared between the two cohorts. Finally, a prediction model for evaluating STAS status in stage IA lung adenocarcinoma was established by a random forest model. Results. Among 92 patients with stage IA lung adenocarcinoma, STAS positive was identified in 19 patients. The random forest classification model identified predictive features, including CT maximum value, consolidation to tumor ratio (CTR), 3D diameter, CT mean value, entropy, and CT minimum value. The misclassification rate of the random forest model is only 7.69. Conclusion. The risk factors of STAS in stage IA lung adenocarcinoma can be effectively identified based on the random forest model, and the hierarchical management of characteristic risk can be effectively realized. A random forest model for predicting STAS in IA lung adenocarcinoma is simple and practical.