Search

Background: Predicting an individual's risk of death from COVID-19 is essential for planning and optimising resources. However, since the real-world mortality rate is relatively low, particularly in places like Hong Kong, this makes building an accurate prediction model difficult due to the imbalanced nature of the dataset. This study introduces an innovative application of graph convolutional networks (GCNs) to predict COVID-19 patient survival using a highly imbalanced dataset. Unlike traditional models, GCNs leverage structural relationships within the data, enhancing predictive accuracy and robustness. By integrating demographic and laboratory data into a GCN framework, our approach addresses class imbalance and demonstrates significant improvements in prediction accuracy. Methods: The cohort included all consecutive positive COVID-19 patients fulfilling study criteria admitted to 42 public hospitals in Hong Kong between January 23 and December 31, 2020 (n = 7,606). We proposed the population-based graph convolutional neural network (GCN) model which took blood test results, age and sex as inputs to predict the survival outcomes. Furthermore, we compared our proposed model to the Cox Proportional Hazard (CPH) model, conventional machine learning models, and oversampling machine learning models. Additionally, a subgroup analysis was performed on the test set in order to acquire a deeper understanding of the relationship between each patient node and its neighbours, revealing possible underlying causes of the inaccurate predictions. Results: The GCN model was the top-performing model, with an AUC of 0.944, considerably outperforming all other models (p < 0.05), including the oversampled CPH model (0.708), linear regression (0.877), Linear Discriminant Analysis (0.860), K-nearest neighbours (0.834), Gaussian predictor (0.745) and support vector machine (0.847). With Kaplan-Meier estimates, the GCN model demonstrated good discriminability between low- and high-risk individuals (p < 0.0001). Based on subanalysis using the weighted-in score, although the GCN model was able to discriminate well between different predicted groups, the separation was inadequate between false negative (FN) and true negative (TN) groups. Conclusion: The GCN model considerably outperformed all other machine learning methods and baseline CPH models. Thus, when applied to this imbalanced COVID survival dataset, adopting a population graph representation may be an approach to achieving good prediction. [ABSTRACT FROM AUTHOR]

Multiparametric magnetic resonance imaging (mpMRI) has emerged as a valuable diagnostic tool in prostate lesion assessment. However, training convolutional neural networks (CNNs) inevitably involves magnetic resonance (MR) images from multiple cohorts. There always exists variation in scanning protocol among cohorts, inducing significant changes in data distribution between source and target domains. This challenge has greatly limited clinical adoption on a large scale. Herein, a coarse mask‐guided deep domain adaptation network (CMD2A‐Net) is proposed to develop a fully automated framework for prostate lesion detection and classification (PLDC). No category or mask label is required from the target domain. A coarse segmentation module is trained to cover the possible lesion‐related regions, so that attention maps can be generated to dedicate the local feature extraction of lesions within those regions. Experiments are performed on 512 mpMRI sets from datasets of PROSTATEx (330 sets) and two cohorts, A (74 sets) and B (108 sets). Using ensemble learning, CMD2A‐Net accomplishes an AUC of 0.921 in cohort A and 0.913 in cohort B, demonstrating its transferability from a large‐scale public dataset PROSTATEx to small‐scale target domains. Results from an ablation study also support its effectiveness in classification between benign and malignant lesions, compared to the state‐of‐the‐art models. An interactive preprint version of the article can be found here: https://doi.org/10.22541/au.166081031.11420810/v1. [ABSTRACT FROM AUTHOR]

This study proposed a semisupervised loss function named level‐set loss (LSLoss) for cerebral white matter hyperintensities (WMHs) segmentation on fluid‐attenuated inversion recovery images. The training procedure did not require manually labeled WMH masks. Our image preprocessing steps included biased field correction, skull stripping, and white matter segmentation. With the proposed LSLoss, we trained a V‐Net using the MRI images from both local and public databases. Local databases were the small vessel disease cohort (HKU‐SVD, n = 360) and the multiple sclerosis cohort (HKU‐MS, n = 20) from our institutional imaging center. Public databases were the Medical Image Computing Computer‐assisted Intervention (MICCAI) WMH challenge database (MICCAI‐WMH, n = 60) and the normal control cohort of the Alzheimer's Disease Neuroimaging Initiative database (ADNI‐CN, n = 15). We achieved an overall dice similarity coefficient (DSC) of 0.81 on the HKU‐SVD testing set (n = 20), DSC = 0.77 on the HKU‐MS testing set (n = 5), and DSC = 0.78 on MICCAI‐WMH testing set (n = 30). The segmentation results obtained by our semisupervised V‐Net were comparable with the supervised methods and outperformed the unsupervised methods in the literature. [ABSTRACT FROM AUTHOR]

Background: Lung cancer is the leading cause of cancer-related mortality, and accurate prediction of patient survival can aid treatment planning and potentially improve outcomes. In this study, we proposed an automated system capable of lung segmentation and survival prediction using graph convolution neural network (GCN) with CT data in non-small cell lung cancer (NSCLC) patients. Methods: In this retrospective study, we segmented 10 parts of the lung CT images and built individual lung graphs as inputs to train a GCN model to predict 5-year overall survival. A Cox proportional-hazard model, a set of machine learning (ML) models, a convolutional neural network based on tumor (Tumor-CNN), and the current TNM staging system were used as comparison. Findings: A total of 1,705 patients (main cohort) and 125 patients (external validation cohort) with lung cancer (stages I and II) were included. The GCNmodel was significantly predictive of 5-year overall survival with an AUC of 0.732 (p < 0.0001). The model stratified patients into low- and high-risk groups, which were associated with overall survival (HR = 5.41; 95% CI:, 2.32-10.14; p < 0.0001). On external validation dataset, our GCN model achieved the AUC score of 0.678 (95% CI: 0.564-0.792; p < 0.0001). Interpretation: The proposed GCN model outperformed all ML, Tumor-CNN, and TNM staging models. This study demonstrated the value of utilizing medical imaging graph structure data, resulting in a robust and effective model for the prediction of survival in early-stage lung cancer. [ABSTRACT FROM AUTHOR]

The study population contains 145 patients who were prospectively recruited for coronary CT angiography (CCTA) and fundoscopy. This study first examined the association between retinal vascular changes and the Coronary Artery Disease Reporting and Data System (CAD-RADS) as assessed on CCTA. Then, we developed a graph neural network (GNN) model for predicting the CAD-RADS as a proxy for coronary artery disease. The CCTA scans were stratified by CAD-RADS scores by expert readers, and the vascular biomarkers were extracted from their fundus images. Association analyses of CAD-RADS scores were performed with patient characteristics, retinal diseases, and quantitative vascular biomarkers. Finally, a GNN model was constructed for the task of predicting the CAD-RADS score compared to traditional machine learning (ML) models. The experimental results showed that a few retinal vascular biomarkers were significantly associated with adverse CAD-RADS scores, which were mainly pertaining to arterial width, arterial angle, venous angle, and fractal dimensions. Additionally, the GNN model achieved a sensitivity, specificity, accuracy and area under the curve of 0.711, 0.697, 0.704 and 0.739, respectively. This performance outperformed the same evaluation metrics obtained from the traditional ML models (p < 0.05). The data suggested that retinal vasculature could be a potential biomarker for atherosclerosis in the coronary artery and that the GNN model could be utilized for accurate prediction. [ABSTRACT FROM AUTHOR]

Neuromodulation techniques such as deep brain stimulation (DBS) are a promising treatment for memory-related disorders including anxiety, addiction, and dementia. However, the outcomes of such treatments appear to be somewhat paradoxical, in that these techniques can both disrupt and enhance memory even when applied to the same brain target. In this article, we hypothesize that disruption and enhancement of memory through neuromodulation can be explained by the dropout of engram nodes. We used a convolutional neural network (CNN) to classify handwritten digits and letters and applied dropout at different stages to simulate DBS effects on engrams. We showed that dropout applied during training improved the accuracy of prediction, whereas dropout applied during testing dramatically decreased the accuracy of prediction, which mimics enhancement and disruption of memory, respectively. We further showed that transfer learning of neural networks with dropout had increased the accuracy and rate of learning. Dropout during training provided a more robust "skeleton" network and, together with transfer learning, mimicked the effects of chronic DBS on memory. Overall, we showed that the dropout of engram nodes is a possible mechanism by which neuromodulation techniques such as DBS can both disrupt and enhance memory, providing a unique perspective on this paradox. [ABSTRACT FROM AUTHOR]

• Deep learning provides efficient and accurate prediction of treatment response. • Transfer learning can be used in radiological task with insufficient image datasets. • Tumor microenvironment and signaling pathways are linked with radiological prediction. Deep learning is promising to predict treatment response. We aimed to evaluate and validate the predictive performance of the CT-based model using deep learning features for predicting pathologic complete response to neoadjuvant chemoradiotherapy (nCRT) in esophageal squamous cell carcinoma (ESCC). Patients were retrospectively enrolled between April 2007 and December 2018 from two institutions. We extracted deep learning features of six pre-trained convolutional neural networks, respectively, from pretreatment CT images in the training cohort (n = 161). Support vector machine was adopted as the classifier. Validation was performed in an external testing cohort (n = 70). We assessed the performance using the area under the receiver operating characteristics curve (AUC) and selected an optimal model, which was compared with a radiomics model developed from the training cohort. A clinical model consisting of clinical factors only was also built for baseline comparison. We further conducted a radiogenomics analysis using gene expression profiles to reveal underlying biology associated with radiological prediction. The optimal model with features extracted from ResNet50 achieved an AUC and accuracy of 0.805 (95% CI, 0.696–0.913) and 77.1% (65.6%-86.3%) in the testing cohort, compared with 0.725 (0.605–0.846)) and 67.1% (54.9%-77.9%) for the radiomics model. All the radiological models showed better predictive performance than the clinical model. Radiogenomics analysis suggested a potential association mainly with WNT signaling pathway and tumor microenvironment. The novel and noninvasive deep learning approach could provide efficient and accurate prediction of treatment response to nCRT in ESCC, and benefit clinical decision making of therapeutic strategy. [ABSTRACT FROM AUTHOR]