Recent advances in deep learning have shown great potential for the automatic generation of medical imaging reports. Deep learning techniques, inspired by image captioning, have made significant progress in the field of diagnostic report generation. This paper provides a comprehensive overview of recent research efforts in deep learning-based medical imaging report generation and proposes future directions in this field. First, we summarize and analyze the data set, architecture, application, and evaluation of deep learning-based medical imaging report generation. Specially, we survey the deep learning architectures used in diagnostic report generation, including hierarchical RNN-based frameworks, attention-based frameworks, and reinforcement learning-based frameworks. In addition, we identify potential challenges and suggest future research directions to support clinical applications and decision-making using medical imaging report generation systems.

Lung segmentation in high-resolution computed tomography (HRCT) images is necessary before the computer-aided diagnosis (CAD) of interstitial lung disease (ILD). Traditional methods are less intelligent and have lower accuracy of segmentation. This paper develops a novel automatic segmentation model using radiomics with a combination of hand-crafted features and deep features. The study uses ILD Database-MedGIFT from 128 patients with 108 annotated image series and selects 1946 regions of interest (ROI) of lung tissue patterns for training and testing. First, images are denoised by Wiener filter. Then, segmentation is performed by fusion of features that are extracted from the gray-level co-occurrence matrix (GLCM) which is a classic texture analysis method and U-Net which is a standard convolutional neural network (CNN). The final experiment result for segmentation in terms of dice similarity coefficient (DSC) is 89.42%, which is comparable to the state-of-the-art methods. The training performance shows the effectiveness for a combination of texture and deep radiomics features in lung segmentation.

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.

The construction of smart campuses can help realize the modernization of higher education. However, the subjects and mechanisms are easily ignored in the construction of smart campuses in colleges and universities. The purpose of this paper is to construct an evolutionary game model based on the complex network structure characteristics of smart campus subjects and the dynamic relationship of collaborative games. Taking the scale-free network as the carrier, the decision-making behavior among subjects is explained. Then, the importance of subject coordination and the effectiveness of the mechanisms of leadership organization, special funds, assessment rewards, and punishments are analyzed through the simulation platform, i.e., Simpy. The results show that (1) moderate subject collaboration helps to avoid the phenomenon of getting something for nothing, (2) appropriate leadership and organization mechanisms can promote the construction process, (3) special fund mechanisms can ensure sustainable development, and (4) assessment reward and punishment mechanisms can increase the popularity of achievements. The proposed method enriches relevant theories of smart campus research and provides a reference for decision-making in smart campus construction.

Generative adversarial network (GAN) synthesizing images has been widely applied to data augmentation. However, existing GAN-based imaging generation has drawbacks in explaining the synthetic principles. Hence, generative approaches resolving the limited medical images for deep learning-based computer aided diagnosis (CAD) which requires large amounts of data are scanty. In this paper, we propose an interpretable data synthesis network using modified triple generative adversarial network deployed into deep learning-based breast imaging reporting and data system (BI-RADS) classification of breast ultrasound mass. The mass synthetic network consists of an interpretable GAN that grasps the standardized characterizations described in BI-RADS for mass. Then diagnostic network with synthetic data determines the BI-RADS level of breast mass based on multi-classifier. Experiments were performed on a private breast ultrasound mass dataset to show that the synthetic network can provide interpretability of generative mass. In addition, we demonstrate that the synthetic masses improve the performance of breast cancer classification.

Introduction: Suicidal ideation (SI) in patients with major depressive disorder (MDD) is frequently underrecognized in early clinical assessment, owing to the complexity of its underlying neurobiological mechanisms and the lack of complementary objective biomarkers. To address this issue, this study proposes an interpretable deep learning framework designed to assist in the diagnosis of patients with MDD with suicidal ideation (MDDSI) and to elucidate its underlying neural mechanisms. Methods: A modified BrainNet convolutional neural network architecture was employed, utilizing a whole-brain functional connectivity (FC) matrix derived from resting-state functional magnetic resonance imaging as input features. A gradient-weighted class activation mapping algorithm was subsequently applied to visualize key brain regions. The study included 356 patients with MDDSI and 107 patients with MDD without suicidal ideation (MDDNSI) as the control group. Results: Five-fold cross-validation indicated that the model achieved an accuracy of 88%, sensitivity of 95%, specificity of 85%, and an area under the curve of 0.93 on the test set. Feature visualization results revealed that the model’s classification decisions primarily relied on abnormal FC patterns in regions such as the motor cortex, anterior cingulate cortex, occipital lobe, temporal lobe, parietal lobe, and cerebellum. Discussion: This work provides a valuable reference for both auxiliary diagnosis and the mechanistic investigation of MDDSI.