Meiyan Liang

Quantitative assessment of gigapixel whole slide images remains challenging due to morphological diversity and human intervention during slide acquisition and processing. While AI-assisted diagnostics hold promise for improving healthcare accessibility and quality, practical application is hindered by difficulties in faithfully representing histomorphological heterogeneity and scaling AI models for deployment. Prevailing approaches predominantly rely on weakly supervised algorithms based on homogeneous graph, which typically decouple diagnostic models from the underlying hardware, resulting in lack of an efficient, closed-loop, end-to-end system. Here, we propose a portable integrated system: the Intelligent Pathology Whole-Slide Analyzer (iPathWS analyzer), which embeds a Spectral Heterogeneity Engine Network and interfaces directly with a slide scanner to form a unified platform for WSI inference. The system seamlessly overlays explainable heatmaps onto slides, enabling intuitive AI integration into routine clinical workflows in a closed-loop manner. We validate the effectiveness of the iPathWS analyzer across four distinct classification tasks spanned by three independent pathology datasets, including multi-center crossover studies. Furthermore, the model produces highly detailed and interpretable heatmaps capable of detecting tiny lesions with the size of 188.5× 280.2μm², which are often imperceptible by human observers. Overall, the iPathWS Analyzer demonstrates consistent and substantial improvements in diagnostic performance, offering a scalable and robust platform for next-generation computer-aided pathology.

Gigapixel whole-slide image (WSI) prediction and region-of-interest localization present considerable challenges due to the diverse range of features both across different slides and within individual slides. Most current methods rely on weakly supervised learning using homogeneous graphs to establish context-aware relevance within slides, often neglecting the rich diversity of heterogeneous information inherent in pathology images. Inspired by the negative sampling strategy of the Determinantal Point Process (DPP) and the hierarchical structure of pathology slides, we introduce the Negative Sample Boosted Hierarchical Heterogeneous Graph Attention Network (NSB-H²GAN). This model addresses the over-smoothing issue typically encountered in classical Graph Convolutional Networks (GCNs) when applied to pathology slides. By incorporating "negative samples" at multiple scales and utilizing hierarchical, heterogeneous feature discrimination, NSB-H²GAN more effectively captures the unique features of each patch, leading to an improved representation of gigapixel WSIs. We evaluated the performance of NSB-H²GAN on three publicly available datasets: CAMELYON16, TCGA-NSCLC and TCGA-COAD. The results show that NSB-H²GAN significantly outperforms existing state-of-the-art methods in both qualitative and quantitative evaluations. Moreover, NSB-H²GAN generates more detailed and interpretable heatmaps, allowing for precise localization of tiny lesions as small as 200 μm × 200μm that are often missed by the human eye. The robust performance of NSB-H²GAN offers a new paradigm for computer-aided pathology diagnosis and holds great potential for advancing the clinical applications of computational pathology.

Recently, the DB algorithm has drawn considerable attention in scene text detection due to its differentiable binarization module, which is proposed to simplify the complex post-processing of the existing segmentation-based scene text detection approaches. However, DB is limited to its layer-wise multiscale feature representation, semantic information loss for the feature map of the highest level, insufficient localization signals at the higher layers, and less scale robustness of the segmentation model. In this paper, we propose a novel scene text detector with multiscale feature augmentation and attention mechanisms (MFAAM). Specifically, Res2Net serves as the backbone network for extracting fine-grained features with multiple scales. In feature fusion, we construct the Deep Feature Enhancement (DFE) module which can extract ratio-invariant spatial context information to reduce the semantic information loss in the top-level feature map. The Feature Pyramid Augmentation (FPA) module is employed to fuse the lower-level positioning information along the bottom-up path by which the higher-level features are strengthened. Furthermore, the improvement for scale robustness of the segmentation model is implemented by Attentional Feature Fusion (AFF), which can adaptively integrate multiscale features through the attention mechanisms with the channel and spatial context. Experiments on the ICDAR2015 dataset validate the superiority of the proposed method. It is worth noting that the presented detector outperforms DB by 4.38 percent on precision, 2.88 percent on recall, and 3.55 percent on F-measure. Compared with the state-of-the-art algorithms, the presented algorithm exhibits a higher F-measure.

Shallow underground explosive source localization technology is a key technology in the field of underground space localization. The existing approaches mainly aim to improve the localization accuracy, but need to deploy enormous sensors in the monitoring area, and rely on a large number of back-end workstations to solve. These methods have the defects of considerable calculation and high time cost, and are hard to satisfy the precise and real-time requirements of onsite testing, ultimately resulting in slow localization speed and accurate localization failure. Fortunately, emerging deep reinforcement learning (DRL) can effectively solve the problem of slow search policy by modeling the source localication as a Markov decision process (MDP). Therefore, a Curiosity-Driven Deep Dueling Double Q-learning Network (C-D3QN) is subsequently proposed to solve the above MDP. The overestimation problem is solved by decoupling selection and evaluation of the bootstrap action, and the action difference is effectively increased by introducing the dueling network that separately represents state values and action advantages. Meanwhile, the exploration is jointly reinforced by an intrinsic reward outputted from the curiosity module and an extrinsic reward supplied by the environment, guaranteeing the convergence to global optimal. Finally, extensive simulation results based on the outfield experiment data show that compared with other algorithms, the proposed scheme can significantly improve exploration ability and learning speed as well as generalization and robustness. Additionally, compared to the baseline algorithm DQN, the C-D3QN algorithm can offer an improved localization accuracy as high as 99.62% and an increased localization speed of 66.23%.

With the rapid development of Internet of things (IoT) and mobile communication technology, Internet of vehicles (IoV) has potential prospects in providing live streaming services and improving the quality of experience (QoE) of ground mobile vehicles (GMVs). However, when the communication link is blocked in emergency situations such as road congestion and base station (BS) paralysis, it is still an arduous task to provide realtime high-definition video for vehicle users. Fortunately, as the aerial mobile BS, unmanned aerial vehicle (UAV) is expected to be deployed for emergency communication recovery in extreme scenarios. But high mobility of GMVs and UAV leads to dynamic changes of network topology, how to establish a stable communication link between UAV and GMVs to ensure the QoE of vehicle users has become a huge challenge. Therefore, this study proposes a novel live video transmission scheme based on scalable video coding (SVC). By jointly optimizing the UAV trajectory, video layer selection, and bandwidth resource allocation, our objective is to maximize long-term QoE while reducing UAV’s flight energy consumption. To maximize QoE, time delay and bitrate switching at GMVs’ receiver are set as penalty items while maximizing the reward of video quality. The joint optimization problem is modeled as a Markov decision process (MDP), and is solved by the soft actor-critic (SAC) deep reinforcement learning (DRL) algorithm, which is based on the maximum entropy framework. Extensive simulation results based on real-world datasets show that compared with other DRL algorithms, the proposed scheme has excellent performance in all respects.

Digital pathology images are treated as the “gold standard” for the diagnosis of colorectal lesions, especially colon cancer. Real-time, objective and accurate inspection results will assist clinicians to choose symptomatic treatment in a timely manner, which is of great significance in clinical medicine. However, Manual methods suffers from long inspection cycle and serious reliance on subjective interpretation. It is also a challenging task for existing computer-aided diagnosis methods to obtain models that are both accurate and interpretable. Models that exhibit high accuracy are always more complex and opaque, while interpretable models may lack the necessary accuracy. Therefore, the framework of ensemble adaptive boosting prototype tree is proposed to predict the colorectal pathology images and provide interpretable inference by visualizing the decision-making process in each base learner. The results showed that the proposed method could effectively address the “accuracy-interpretability trade-off” issue by ensemble of m adaptive boosting neural prototype trees. The superior performance of the framework provides a novel paradigm for interpretable inference and high-precision prediction of pathology image patches in computational pathology.