Dementia is characterized by progressive neurodegeneration that unfolds heterogeneously across brain regions and cognitive stages. While hippocampal atrophy has traditionally dominated biomarker research, mounting evidence suggests that earlier and region-specific cortical degeneration may precede classical markers. In this study, we perform a region-wise voxel-based MRI analysis across NonDemented, VeryMildDemented (VMD), and MildDemented (MD) stages to address three critical questions: (1) What brain region serves as a biomarker of transition? (2) What is the rate of progression? (3) What drives early disease? We identify the left inferior parietal cortex (Parietal_Inf_L, Region_65) as a key early biomarker, showing 5.5% atrophy in VMD and a sharp 100% decline in MD ($p<1 \times 10^{-30}$), with a steep regression slope (−21.3). Disease progression follows a nonlinear trajectory, with the posterior parietal cortex-including the precuneus and superior/inferior parietal lobules-emerging as the main driver of early degeneration. Notably, the hippocampus exhibited no measurable signal, emphasizing the need to reassess its role in early-stage imaging. Our findings establish posterior parietal atrophy as both an early biomarker and a core substrate of dementia progression, offering new anatomical targets for early detection and intervention.

In medical image segmentation across multiple modalities (e.g., MRI, CT, etc.) and heterogeneous data sources (e.g., different hospitals and devices), Domain Generalization (DG) remains a critical challenge in AI-driven healthcare. This challenge primarily arises from domain shifts, imaging variations, and patient diversity, which often lead to degraded model performance in unseen domains. To address these limitations, we identify key issues in existing methods, including insufficient simplification of complex style features, inadequate reuse of domain knowledge, and a lack of feedback-driven optimization. To tackle these problems, inspired by Feynman’s learning techniques in educational psychology, this paper introduces a cognitive science-inspired meta-learning paradigm for medical image domain generalization segmentation. We propose, for the first time, a cognitive-inspired Feynman-Guided Meta-Learning framework for medical image domain generalization segmentation (FGML-DG), which mimics human cognitive learning processes to enhance model learning and knowledge transfer. Specifically, we first leverage the ‘concept understanding’ principle from Feynman’s learning method to simplify complex features across domains into style information statistics, achieving precise style feature alignment. Second, we design a meta-style memory and recall method (MetaStyle) to emulate the human memory system’s utilization of past knowledge. Finally, we incorporate a Feedback-Driven Re-Training strategy (FDRT), which mimics Feynman’s emphasis on targeted relearning, enabling the model to dynamically adjust learning focus based on prediction errors. Experimental results demonstrate that our method outperforms other existing domain generalization approaches on two challenging medical image domain generalization tasks.

Camouflaged Object Detection is challenging due to the high degree of similarity between camouflaged objects and their surrounding backgrounds. Current COD methods mainly rely on edge extraction in the spatial domain and local pixel-level information, neglecting the importance of global structural features. Additionally, they fail to effectively leverage the importance of phase spectrum information within frequency domain features. To this end, we propose a COD framework BASFNet based on boundary-aware frequency domain and spatial domain fusion. This method uses dual guided integration of frequency domain and spatial domain features. A phase-spectrum-based frequency-enhanced edge exploration module (FEEM) and a spatial core segmentation module (SCSM) are introduced to jointly capture the boundary and object features of camouflaged objects. These features are then effectively integrated through a spatial-frequency fusion interaction module (SFFIM). Furthermore, the boundary detection is further optimized through an boundary-aware training strategy. BASFNet outperforms existing state-of-the-art methods on three benchmark datasets, validating the effectiveness of the fusion of frequency and spatial domain information in COD tasks.

Ancient character detection plays a pivotal role in preserving historical scripts and advancing archaeological research. However, the unique characteristics of ancient texts—such as incomplete forms, fragmented strokes, and irregular backgrounds—make it challenging for conventional text detection methods to succeed. To address these challenges, we introduce CharDetNet, a lightweight and highly efficient model tailored for ancient character detection. Built upon the robust YOLO architecture, CharDetNet incorporates two key innovations: StrokeEnhance Attention (SE-ATT) and RegionFocus Attention (RF-ATT). The SE-ATT module refines the detection of subtle stroke details by leveraging high-frequency features, compensating for the damage or blending caused by ancient scanning techniques. Meanwhile, the RF-ATT module utilizes spatial context to suppress irrelevant background noise, focusing the model’s attention on the character regions. To further enhance performance under limited training data, we introduce ClusterRefine Score (CR-Score), a post-processing algorithm based on spectral clustering that boosts prediction recall. Extensive experiments on two challenging datasets—Chinese oracle bone script and Egyptian hieroglyphs—demonstrate that CharDetNet surpasses existing detection models, achieving a significant increase in AP by 9.8-12.5%. Our method sets a new standard in ancient character detection, offering both high accuracy and computational efficiency. The implementation will be available on GitHub.

Deep learning models in medical image segmentation and classification tasks can automatically extract features and perform high-performance inference, thereby assisting doctors in achieving efficient and accurate automated decision support. However, these models are typically trained for a single task, which leads to limitations such as ignoring task relevance and poor scalability. In response to the above problems, we hope to apply one model to different tasks to enhance the model’s scalability. Hence, this paper proposes an Alternating Multi-Task Hierarchical Network (AMTH-Net) for medical image segmentation and classification. The model is divided into three hierarchical modules: Pathological Region Clarity (PRC) serves as an auxiliary module to improve segmentation and classification capabilities, the Multi-Resolution Attention (MRA) segmentation module focuses on image information at different resolution levels through deep supervision to enhance segmentation accuracy, and the Cascading Multi-Scale Information (CMSI) classification module employs a cascading multi-scale mechanism to gradually integrate discrete information from different network layers, thereby enhancing classification performance. Additionally, we propose a novel Alternating Interaction Loss (AI-Loss) based on a Multi-Level Gradient Information Feedback (MGIF) algorithm to further improve the model’s segmentation and diagnostic performance. We validated our approach on two datasets: the public COVID CXR dataset and our newly proposed F_BUSI breast ultrasound dataset. Experimental results demonstrate that AMTH-Net achieves excellent performance in both segmentation and classification, and it outperforms existing methods in terms of segmentation and classification capabilities.

Military-civilian attribute recognition of ships in synthetic aperture radar (SAR) imagery plays an important role in marine surveillance. However, high-quality labeled data are hard to obtain for SAR ships, which hinder the development of deep learning models. Considering that models directly transferred from labeled optical images cannot achieve satisfactory performance for SAR applications due to the great discrepancy of different modalities, we propose a two-stage transfer learning method by combining the data-level and feature-level knowledge transfer. First, CycleGAN is adopted in the first stage to transfer the labeled optical image domain to the intermediate SAR-like image domain with the attribute labels. Then, a novel network called Domain Transfer using Adversarial learning and Metric learning (DTAM) is proposed to realize the task of military-civilian ship recognition by the domain adaption of the intermediate domain and the target SAR domain with joint adversarial learning and metric learning. To validate the proposed method, we establish a high-resolution SAR ship recognition dataset (HRSSRD), containing SAR and optical images of military and civilian ships. The experimental results show that the proposed two-stage architecture exhibits promising performance on the problem of SAR military-civilian ship recognition.