Various retinal vessel segmentation methods based on convolutional neural networks were proposed recently, and Dense U-net as a new semantic segmentation network was successfully applied to scene segmentation. Retinal vessel is tiny, and the features of retinal vessel can be learned effectively by the patch-based learning strategy. In this study, we proposed a new retinal vessel segmentation framework based on Dense U-net and the patch-based learning strategy. In the process of training, training patches were obtained by random extraction strategy, Dense U-net was adopted as a training network, and random transformation was used as a data augmentation strategy. In the process of testing, test images were divided into image patches, test patches were predicted by training model, and the segmentation result can be reconstructed by overlapping-patches sequential reconstruction strategy. This proposed method was applied to public datasets DRIVE and STARE, and retinal vessel segmentation was performed. Sensitivity (Se), specificity (Sp), accuracy (Acc), and area under each curve (AUC) were adopted as evaluation metrics to verify the effectiveness of proposed method. Compared with state-of-the-art methods including the unsupervised, supervised, and convolutional neural network (CNN) methods, the result demonstrated that our approach is competitive in these evaluation metrics. This method can obtain a better segmentation result than specialists, and has clinical application value.

Motor imagery electroencephalogram (MI-EEG) classification is essential in brain-computer interface (BCI), and many classification methods have been proposed recently. However, the MI-EEG classification accuracy of the public dataset still has room for improvement, and designing a suitable model to extract and fuse the multi-modality features efficiently is crucial. In this study, we proposed a Multi-scale Space-time Frequency fusion Network (MSFNet) to improve the MI-EEG classification accuracy. The MSFNet comprises data acquisition and preprocessing, multi-scale time-conv fusion unit, multi-scale frequency-conv fusion unit, feature fusion, and classification. Multi-scale time-conv fusion unit can extract multi-scale spatiotemporal features, and multi-scale frequency-conv fusion unit can extract five frequency sub-band features. These two features were concatenated to complete the multi-modality features fusion, and MI-EEG was classified. Average accuracy, kappa value, and F1 score were adopted as the evaluation metric, and BCI Competition 2008 IV 2a and High Gamma datasets were employed to demonstrate the effectiveness of the MSFNet. The superiority of this proposed model was demonstrated by comparison against the state-of-the-art methods, and the classification result is the highest. Overall, we achieved an average accuracy of 80.47%, kappa value of 0.783, and F1 score of 0.743 in the BCI Competition 2008 IV 2a dataset, and we achieved an average accuracy of 93.56%, kappa value of 0.933, and F1 score of 0.915 in High Gamma dataset. This model realized the extraction and efficient fusion of spatiotemporal and frequency domain features and the high-precision MI-EEG classification, which has an important application value.

Overexpression of lysine specific demethylase 1 (LSD1) has been found in many cancers. New anticancer drugs targeting LSD1 have been designed. The research on irreversible LSD1 inhibitors has entered the clinical stage, while the research on reversible LSD1 inhibitors has progressed slowly so far. In this study, 41 stilbene derivatives were studied as reversible inhibitors by three-dimensional quantitative structure–activity relationship (3D-QSAR). Comparative molecular field analysis (CoMFA ${\mathrm{q}}^2$ = 0.623, ${\mathrm{r}}^2$ = 0.987, ${\mathrm{r}}_{\mathrm{pred}}^2$ = 0.857) and comparative molecular similarity indices analysis (CoMSIA ${\mathrm{q}}^2$ = 0.728, ${\mathrm{r}}^2$ = 0.960, ${\mathrm{r}}_{\mathrm{pred}}^2$ = 0.899) were used to establish the model, and the structure–activity relationship of the compounds was explained by the contour maps. The binding site was predicted by two different kinds of software, and the binding modes of the compounds were further explored. A series of key amino acids Val288, Ser289, Gly314, Thr624, Lys661 were found to play a key role in the activity of the compounds. Molecular dynamics (MD) simulations were carried out for compounds 04, 17, 21, and 35, which had different activities. The reasons for the activity differences were explained by the interaction between compounds and LSD1. The binding free energy was calculated by molecular mechanics generalized Born surface area (MM/GBSA). We hope that this research will provide valuable information for the design of new reversible LSD1 inhibitors in the future.

Diffeomorphic demons can not only guarantee smooth and reversible deformation, but also avoid unreasonable deformation. However, the number of iterations which has great influence on the registration result needs to be set manually. This study proposed a novel method to exploit the adaptive diffeomorphic multi-resolution demons algorithm to the non-rigid registration of the same modality medical images with large deformation. Firstly an optimized non-rigid registration framework and the diffeomorphism strategy were used, and then a similarity energy function based on the grey value was designed as registration metric, lastly termination condition was set based on the variation of this metric and iterations can be stopped adaptively. Quantitative analyses based on the registration evaluation indexes were conducted to prove the validity of this method. Registration result of synthetic image and the same modality MRI and CT image was compared with those obtained by other demons algorithms. Quantitative analyses demonstrated the proposed method’s superiority. Medical image with large deformation was produced by rotational distortion and extrusion transform, and the same modality image registration with large deformation was performed successfully. Quantitative analyses showed that the registration evaluation indexes remained stable with an increase in transform strength. This method can be also applied to pulmonary medical image registration with large deformation successfully, and it showed the clinical application value. The influence of different driving forces and parameters on the registration result was analysed, and the result demonstrated that the proposed method is effective and robust. This method can solve the non-rigid registration problem of the same modality medical image with large deformation showing promise for diagnostic pulmonary imaging applications.

Objective. Generally, due to a lack of explainability, radiomics based on deep learning has been perceived as a black-box solution for radiologists. Automatic generation of diagnostic reports is a semantic approach to enhance the explanation of deep learning radiomics (DLR). Approach. In this paper, we propose a novel model called radiomics-reporting network (Radioport), which incorporates text attention. This model aims to improve the interpretability of DLR in mammographic calcification diagnosis. Firstly, it employs convolutional neural networks to extract visual features as radiomics for multi-category classification based on breast imaging reporting and data system. Then, it builds a mapping between these visual features and textual features to generate diagnostic reports, incorporating an attention module for improved clarity. Main results. To demonstrate the effectiveness of our proposed model, we conducted experiments on a breast calcification dataset comprising mammograms and diagnostic reports. The results demonstrate that our model can: (i) semantically enhance the interpretability of DLR; and, (ii) improve the readability of generated medical reports. Significance. Our interpretable textual model can explicitly simulate the mammographic calcification diagnosis process.