Attention-based generative adversarial network in medical imaging: A narrative review.

As a popular probabilistic generative model, generative adversarial network (GAN) has been successfully used not only in natural image processing, but also in medical image analysis and computer-aided diagnosis. Despite the various advantages, the applications of GAN in medical image analysis face new challenges. The introduction of attention mechanisms, which resemble the human visual system that focuses on the task-related local image area for certain information extraction, has drawn increasing interest. Recently proposed transformer-based architectures that leverage self-attention mechanism encode long-range dependencies and learn representations that are highly expressive. This motivates us to summarize the applications of using transformer-based GAN for medical image analysis. We reviewed recent advances in techniques combining various attention modules with different adversarial training schemes, and their applications in medical segmentation, synthesis and detection. Several recent studies have shown that attention modules can be effectively incorporated into a GAN model in detecting lesion areas and extracting diagnosis-related feature information precisely, thus providing a useful tool for medical image processing and diagnosis. This review indicates that research on the medical imaging analysis of GAN and attention mechanisms is still at an early stage despite the great potential. We highlight the attention-based generative adversarial network is an efficient and promising computational model advancing future research and applications in medical image analysis.

Attention-Based Multi-Scale Generative Adversarial Network for synthesizing contrast-enhanced MRI.

In clinical practice, about 35% of MRI scans are enhanced with Gadolinium - based contrast agents (GBCAs) worldwide currently. Injecting GBCAs can make the lesions much more visible on contrast-enhanced scans. However, the injection of GBCAs is high-risk, time-consuming, and expensive. Utilizing a generative model such as an adversarial network (GAN) to synthesize the contrast-enhanced MRI without injection of GBCAs becomes a very promising alternative method. Due to the different features of the lesions in contrast-enhanced images while the single-scale feature extraction capabilities of the traditional GAN, we propose a new generative model that a multi-scale strategy is used in the GAN to extract different scale features of the lesions. Moreover, an attention mechanism is also added in our model to learn important features automatically from all scales for better feature aggregation. We name our proposed network with an attention-based multi-scale contrasted-enhanced-image generative adversarial network (AMCGAN). We examine our proposed AMCGAN on a private dataset from 382 ankylosing spondylitis subjects. The result shows our proposed network can achieve state-of-the-art in both visual evaluations and quantitative evaluations than traditional adversarial training.Clinical Relevance-This study provides a safe, convenient, and inexpensive tool for the clinical practices to get contrast-enhanced MRI without injection of GBCAs.

Deep Learning-Based Prediction of Future Extrahepatic Metastasis and Macrovascular Invasion in Hepatocellular Carcinoma.

PURPOSE: For timely treatment of extrahepatic metastasis and macrovascular invasion (aggressive progressive disease [PD]) in hepatocellular carcinoma, models aimed at stratifying the risks of subsequent aggressive PD should be constructed. PATIENTS AND METHODS: After dividing 332 patients from five hospitals into training (n = 236) and validation (n = 96) datasets, non-invasive models, including clinical/semantic factors (ModelCS), deep learning radiomics (ModelD), and both (ModelCSD), were constructed to stratify patients according to the risk of aggressive PD. We examined the discrimination and calibration; similarly, we plotted a decision curve and devised a nomogram. Furthermore, we performed analyses of subgroups who received different treatments or those in different disease stages and compared time to aggressive PD and overall survival in the high- and low-risk subgroups. RESULTS: Among the constructed models, ModelCSD, combining clinical/semantic factors and deep learning radiomics, outperformed ModelCS and ModelD (areas under the curve [AUCs] for the training dataset: 0.741, 0.815, and 0.856; validation dataset: 0.780, 0.836, and 0.862), with statistical difference per the net reclassification improvement, the integrated discrimination improvement, and/or the DeLong test in both datasets. Besides, ModelCSD had the best calibration and decision curves. The performance of ModelCSD was not affected by treatment types (AUC: resection = 0.839; transarterial chemoembolization = 0.895; p = 0.183) or disease stages (AUC: BCLC [Barcelona Clinic Liver Cancer] stage 0 and A = 0.827; BCLC stage AB &B = 0.861; p = 0.537). Moreover, the high-risk group had a significantly shorter median time to aggressive PD than the low-risk group (training dataset hazard ratio [HR] = 0.108, p < 0.001; validation dataset HR = 0.058, p < 0.001) and poorer overall survival (training dataset HR = 0.357, p < 0.001; validation dataset HR = 0.204, p < 0.001). CONCLUSION: Our deep learning-based model successfully stratified the risks of aggressive PD. In the high-risk population, current guideline indicates that first-line treatments are insufficient to prevent extrahepatic metastasis and macrovascular invasion and ensure survival benefits, so more therapies may be explored for these patients.

Gene signatures predict biochemical recurrence-free survival in primary prostate cancer patients after radical therapy.

BACKGROUND: This study evaluated the predictive value of gene signatures for biochemical recurrence (BCR) in primary prostate cancer (PCa) patients. METHODS: Clinical features and gene expression profiles of PCa patients were attained from Gene Expression Omnibus (GEO) and The Cancer Genome Atlas (TCGA) datasets, which were further classified into a training set (n = 419), a validation set (n = 403). The least absolute shrinkage and selection operator Cox (LASSO-Cox) method was used to select discriminative gene signatures in training set for biochemical recurrence-free survival (BCRFS). Selected gene signatures established a risk score system. Univariate and multivariate analyses of prognostic factors about BCRFS were performed using the Cox proportional hazards regression models. A nomogram based on multivariate analysis was plotted to facilitate clinical application. Kyoto Encyclopedia of Gene and Genomes (KEGG) and Gene Ontology (GO) analyses were then executed for differentially expressed genes (DEGs). RESULTS: Notably, the risk score could significantly identify BCRFS by time-dependent receiver operating characteristic (t-ROC) curves in the training set (3-year area under the curve (AUC) = 0.820, 5-year AUC = 0.809) and the validation set (3-year AUC = 0.723, 5-year AUC = 0.733). CONCLUSIONS: Clinically, the nomogram model, which incorporates Gleason score and the risk score, could effectively predict BCRFS and potentially be utilized as a useful tool for the screening of BCRFS in PCa.