Past medical research has shown that infantile movement and early neurological development are closely linked. Fidgety Movements that are reflex-like movement occurring in healthy infants less than 20-week of age have proven to be especially important, as past studies have highlighted that their absence is strongly correlated with the future development of neurological disorders like Cerebral Palsy. To provide a timely intervention, the General Movement Assessment was proposed as a screening medical procedure carried out by clinical personnel specifically trained to recognize Fidgety Movements. Because of its high cost in time and resources, several initiatives to automatize General Movement Assessment using machine learning techniques have been proposed in the literature. However none has managed to emerge as state-of-the-art so far. To investigate this problem, we conducted a study using deep learning approaches to learn disentangled feature representations for the recognition of Fidgety Movements using RGB-D video and Inertial Measurement Unit data acquired from 95 infants (average age: $$13.79 \pm 1.40$$ weeks). Our results show that while it is possible to learn features that characterize movement independently of subject information, obtaining feature representations that consistently generalize to subjects unseen during training remains challenging. More specifically, we observe that both the vision- and sensor-based modalities have specific challenges to be addressed for the recognition of Fidgety Movements. We discuss them and provide recommendations to help researchers interested in investigating this problem in the future.

In medicine, the laborious annotation process poses a significant limitation to the creation of large labelled datasets, a challenge that is particularly pronounced for medical video data, where numerous frames require expert annotation. Weakly supervised approaches, such as Multiple Instance Learning (MIL), offer a promising solution by leveraging only video-level labels rather than frame-by-frame annotations. In this work, we tailor a top-k MIL-based approach to medical video data to detect and localise facial motor tics in videos of patients diagnosed with Gilles de la Tourette Syndrome (GTS). Our adaptations include a score-driven top-k instance selection strategy that accounts for variability in tic occurrence between subjects. This enables the effective exploitation of large unlabelled datasets by incorporating weak labels where available. Evaluated against state-of-the-art MIL models, the proposed approach achieves the highest instance-level tic detection performance, with an area under the curve of 86.73

Signal quality assessment is essential for biomedical signal processing, analysis, and interpretation. Various methods exist, including averaged numerical values, thresholding, time-or frequency-domain analysis, and nonlinear approaches. The aim of this study was to evaluate the quality of electrocardiographic (ECG) signals, seismocardiographic signals (SCG), and gyrocardiograms (GCG) based on symmetric projection attractor reconstruction (SPAR) with Takens delay coordinates with fit five classifiers: random forest, gradient boosting, random forest XGB, and support vector machines (SVM) with various number of decision tree-based estimators (100-10,000) and various kernels (linear, radial base function, and polynomial), respectively. The analysis was carried out on a public dataset "Mechanocardiograms with ECG reference" containing 29 concurrent ECG, SCG, and GCG signal recordings. The highest values without SMOTE were observed for ECG signals, SVM with fourth order polynomial kernel (accuracy of 0.6897, PPV of 0.6019, sensitivity of 0.5306, and F1 score of 0.4952), and after applying SMOTE were observed for Gradient Boosting in ECG signal (200 estimators, accuracy 0.7500, PPV of 0.7747, sensitivity of 0.7500, and F2 score of 0.7747 respectively). These findings suggest that the SPAR-based approach is a promising method to accurately assess the quality of cardiovascular signals, including seismocardiograms and gyrocardiograms.

Purpose: Human Activity Recognition (HAR) is a critical application of Artificial Intelligence (AI) and computer vision, aiming to enable machines to comprehend human actions. Despite significant advances in deep learning, HAR faces unique challenges due to video data dimensionality, complex temporal dynamics, and the need to disentangle motion from background noise. We argue that progress is fundamentally driven by the integration of prior knowledge to constrain the learning process. Methods: Unlike conventional reviews that categorize methods based on network architectures, such as Convolutional Neural Networks (CNNs), or input modalities, this work presents a comprehensive survey through a novel, top-down taxonomy organized by the source of prior knowledge. We systematically analyze approaches leveraging priors derived from temporal processing, human kinematics, and spatial relationships, alongside the recent emergence of Large Language Models (LLMs) and Vision-Language Models (VLMs). Results: Our analysis reveals that effective temporal modeling relies on implicit priors (inductive biases) regarding sequential state evolution, whereas robustness to environmental variations is achieved through explicit priors, such as kinematic constraints, skeletal representations and spatial relationships. Furthermore, VLMs are identified as powerful aggregators of common-sense priors, capable of unifying visual features with semantic reasoning. Conclusion: We conclude that the trajectory of HAR is defined not merely by architectural adaptation, but by the sophisticated injection of domain-specific knowledge. This survey provides a unified framework for understanding the interaction between implicit and explicit priors, identifying the integration of common-sense reasoning as a promising avenue for future AI-driven activity understanding.

Since the analysis of erythrocytes and spirochaetes is highly relevant to human health, their automated detection is of significant importance in both medical and computer vision research. However, publicly available datasets in this domain remain scarce. To address this gap, we present the Darkfield Microscopy Image Dataset for Erythrocytes and Spirochaete Detection (DMID-ESD), which consists of 11,794 fully annotated images with labels containing categorical and localization information. We perform comprehensive benchmarking experiments on DMID-ESD to evaluate its utility in tasks such as image classification, object detection, and feature extraction. The results demonstrate that the dataset serves as an effective benchmark for method evaluation. The DMID-ESD dataset is freely available for non-commercial use at: https://figshare. com/articles/dataset/DMID-ESD_zip/22179311.

Time Series Forecasting (TSF) is essential across diverse domains, including finance, supply chain management, environmental monitoring, and healthcare, where reliable predictions support critical decision making. Despite recent progress in Deep Learning (DL) based forecasting, limited data availability and distributional variability often hinder model generalization. Data augmentation has emerged as an effective strategy to mitigate these challenges by increasing data diversity. Existing augmentation methods, such as frequency domain and wavelet based transformations, have shown effectiveness to some extent but struggle to maintain temporal coherence and consistency across varying prediction horizons. To address these limitations, this study proposes WaveFreqAug, a novel hybrid augmentation method that integrates wavelet decomposition with adaptive frequency domain enhancement. The method operates on residual components to enrich spectral variability while preserving long term temporal structures. Comprehensive experiments conducted on four benchmark datasets (ETTh1, ETTh2, ILI, and Weather) using multiple forecasting models demonstrate that WaveFreqAug consistently outperforms baseline and State-of-the-Art (SOTA) augmentation methods in both short and long term forecasting scenarios. These results highlight the robustness and generalizability of the proposed approach for improving deep learning based TSF.

Background From lung cancer and heart disease to rare disorders, research on almost every disease is speeding up. Microscopic cell image analysis is an important area of medical research. As the first step in analysis, segmentation is a significant clinical concern. However, many complexities, such as variations in cell size or shape, overlapping regions, potential poor contrast, and background noise, make automated segmentation of microscopic images a complicated problem. Moreover, there is a notable deficiency in image processing systems that are both user-friendly and capable of delivering credible results.Methods This paper proposed a microscopic cell processing platform to enable efficient and accurate microscopic image analysis. First, the 2018 Data Science Bowl (DSB2018) dataset from the cell segmentation competition of Kaggle in 2018 is grouped into training, validation, and test sets. Then, U-Net++ incorporated with the watershed algorithm was used for microscopic cell image segmentation tasks. Third, a graphics user interface based on QT was designed to display the segmentation process, fine-tune the model according to clinical needs, and automatically generate diagnostic conclusions.Results The average of intersection over union, precision, recall, and F1-score achieved 0.846, 0.908, 0.925, and 0.917, respectively, which were highly satisfactory results, indicating the efficacy of this platform. Moreover, the mean error of the watershed algorithm achieved 0.113, a margin acceptable in clinical diagnostics. Compared with traditional methods, the proposed method significantly improved the performance.Conclusion With the efficient segmentation based on deep learning and accurate quantitative analysis for the results, this microscopic image processing platform may outperform many existing medical analysis systems and have applications in the field of auxiliary medical diagnosis.

Chest X-ray (CXR) analysis models deployed in clinical environments must continuously adapt to newly defined or emerging disease categories, while historical patient images are often unavailable due to privacy, storage, and regulatory constraints. This poses a practical challenge for class-incremental learning (CIL): models should learn new classes without replaying old data while preserving previously acquired diagnostic knowledge. Existing replay-based CIL methods are difficult to apply in privacy-sensitive medical scenarios, whereas vision-only replay-free methods lack stable semantic anchors and are prone to representation drift and catastrophic forgetting. To address this challenge, we propose language-guided knowledge-enhanced class-incremental learning (LK-CIL), a replay-free vision-language framework for medical image classification. LK-CIL freezes a pretrained vision transformer and introduces lightweight task adapters for parameter-efficient incremental updates. To provide stable semantic guidance, it constructs class textual prototypes from clinically reviewed LLM-generated radiographic descriptions. These prototypes guide knowledge-aware adapter merging across tasks, aligning evolving visual representations with class-level linguistic priors. During inference, LK-CIL refines class prototypes using a graph convolutional network on a class-relation graph, selects the most suitable adapter based on feature-prototype similarity, and applies a self-refined prediction strategy to improve robustness. Experiments on ChestX-ray14 show that LK-CIL achieves a favorable balance between recognition accuracy and forgetting suppression compared with state-of-the-art (SOTA) CIL methods under multiple incremental and distribution settings. Additional evaluations further support its effectiveness on an emerging COVID-19 class, multi-label CXR classification, external CXR datasets, and other medical imaging modalities, including colon histopathology and dermoscopy. These results suggest that LK-CIL provides a practical replay-free solution for continuously updating medical image classification models without storing historical patient data. Code is available at: https://github.com/Lingling-Yuan/LK-CIL.

Brain aging is an inevitable process in adulthood, yet there remains a critical need for objective and accurate biomarkers to assess its progression. In this study, we develop a deep learning-based framework for brain age estimation using multiparameter MRI. Structural (T1 and T2 weighted) and diffusion-weighted images were acquired, from which we extracted cortical features, including gray matter volume, surface area, and thickness along with white matter integrity metrics such as fractional anisotropy, mean diffusivity, axial diffusivity, and radial diffusivity. To integrate these multimodal neuroimaging features, we propose MN-FNet (multimodal neurofeature fusion network), a dedicated regression architecture that effectively combines gray matter structure and white matter microstructure. Our model achieves accurate brain age prediction with low estimation error and identifies key neuroanatomical regions associated with aging, additionally providing evidence of hemispheric lateralization as a factor in brain aging. This approach offers a reliable and interpretable tool for brain age estimation, with potential applications in early detection of neurodegenerative conditions.

Colorectal cancer (CRC) screening and diagnosis rely on histopathological assessment, but many high-performing deep learning (DL) models remain computationally demanding and difficult to interpret. This study presents an explainable lightweight receptive field-based convolutional neural network (LRF-CNN) for colorectal histopathological image classification. LRF-CNN integrates multi-branch lightweight receptive field-based (LRF) blocks with an attention block, together with post hoc activation quantification and multi-stage feature-map visualization to improve explainability. On a public five-class colorectal histopathological image dataset, LRF-CNN achieved 92.09% accuracy and outperformed 26 comparative models. The model also showed strong performance on gastric histopathology classification, programmed death-ligand 1 (PD-L1) status prediction in esophageal cancer, and an additional colorectal dataset, with robustness supported by external validation under domain shift. These findings support the LRF-CNN model as an accurate, efficient, and explainable tool for digital pathology.

Pathological image analysis is a key component of digital pathology. The integration of artificial intelligence (AI) has substantially enhanced diagnostic efficiency. However, these models often lack transparency, and their black-box nature limits clinical trust and broader adoption. Although existing reviews have explored the application of explainable AI (XAI) in pathology, most use incomplete or inconsistent taxonomies and lack evaluation frameworks that consider multiple dimensions. The lack of a clear taxonomy and evaluation framework makes it difficult to compare results, identify suitable approaches, and assess the strengths and limitations of XAI methods from multiple perspectives. To address these gaps, this review presents a taxonomy from five perspectives and introduces the PathXAI-2 L minimal evaluation framework. Specifically, it provides a comprehensive overview of key tasks in pathological image analysis, commonly used datasets and their biases, and preprocessing strategies, algorithmic frameworks, and evaluation metrics. Based on this foundation, it proposes a five-dimensional taxonomy of XAI methods, organized by model dependency, explanation scope, explanation stage, explanatory modality, and evidential strength. The evidence perspective is used to distinguish between correlation and causality, by grouping explanations into association, sensitivity, and intervention levels. This review further examines two important factors influencing explainability results: the sensitivity of XAI methods to dataset choice, and the role of external resources in supporting explanations. At the technical level, this review groups XAI methods into seven representative types and introduces XAI tools in pathology. Finally, it proposes a minimal evaluation framework called PathXAI-2 L, focused on the functional and human layers, while highlighting key challenges and future directions for application-grounded evaluation. By integrating tasks, datasets, methods, and evaluation strategies, this review provides a comprehensive reference for advancing both technical progress and clinical impact in pathology XAI.

Few-shot Class-incremental Pill Recognition (FSCIPR) aims to develop an automatic pill recognition system that requires only a few training data and can continuously adapt to new classes, providing technical support for applications in hospitals, portable apps, and assistance for visually impaired individuals. This task faces three core challenges: overfitting, fine-grained classification problems, and catastrophic forgetting. We propose the Well-Prepared Few-shot Class-incremental Learning (WP-FSCIL) framework, which addresses overfitting through a parameter-freezing strategy, enhances the robustness and discriminative power of backbone features with Center-Triplet (CT) loss and supervised contrastive loss for fine-grained classification, and alleviates catastrophic forgetting using a multi-dimensional Knowledge Distillation (KD) strategy based on flexible Pseudo-feature Synthesis (PFS). By flexibly synthesizing any number of old-class features, the PFS strategy resolves the issue of insufficient samples in the KD process, enabling Response-based KD (KD1) and Relation-based KD (KD2) to comprehensively preserve old knowledge. The effectiveness of WP-FSCIL has been validated through experiments conducted on two publicly available pill datasets. These experiments show that WP-FSCIL outperforms existing state-of-the-art methods, demonstrating its superior performance.

Objective assessment of melasma is a critical indicator for evaluating treatment efficacy. However, current clinical practice primarily relies on subjective physician judgment for assessment, which results in significant inter-rater variability. Establishing a reliable automated segmentation method for melasma is essential to achieve precise assessment, and this requires support from high-quality datasets. The Melasma Image Dataset (MEMI-DS) was constructed to address this limitation, comprising 716 images with corresponding annotation files, for training melasma segmentation models. To validate the effectiveness of the dataset, 10 segmentation models (including UNet, UNet++, DeepLabV3, HRNet, DeepLabV3+, LinkNet, MANet, PAN, TransUet, and SegFormer) were systematically evaluated. Experimental results demonstrated that MEMI-DS can effectively distinguish the segmentation performance of different models. Notably, the segmentation performance of models represented by SegFormer was significantly improved by applying data augmentation techniques. These findings not only confirm the effectiveness and practicality of the dataset but also provide reliable technical support for the objective assessment of melasma.

Precise tumor localization and sub-region identification are critical for disease diagnosis. However, current Weakly Supervised Semantic Segmentation (WSSS) methods for brain tumor segmentation are primarily single-class, neglecting differences between tumor sub-regions. We observed that when mainstream transformer-based WSSS methods are applied to multi-class brain tumor segmentation, they encounter two major challenges: sub-region discrimination errors and over-segmentation of small lesions. To address these challenges and advance multi-class WSSS methods for brain tumor analysis, this paper proposes Channel-gated Transformers with Affinity CAM (CTAC). CTAC first employs channel-gated multi-head self-attention to overcome the over-smoothing tendency of the transformer, thereby enhancing inter-class discriminability and improving the model's subclass differentiation capability. Then, CTAC uses multi-scale smoothed affinity to adaptively suppress low-confidence responses in the Class Activation Map (CAM), mitigating over-activation in the CAM, and alleviating the over-segmentation phenomena of small lesions. The proposed CTAC significantly outperformed the baseline method on the BraTS2021 glioma and BraTS2023-MEN meningioma datasets. On Brats2021, it achieved a multi-class mean IoU (mIoU) of 61.718%, an increase of 4.964 percentage points (pp), with the whole-tumor mIoU reaching 79.798% (+6.882 pp). On Brats2023-MEN, CTAC attained 72.887% mIoU (+4.676 pp) for multi-class segmentation and 75.394% (+7.839 pp) for whole-tumor. Furthermore, CTAC surpasses recent state-of-the-art methods. Code is available at https://github.com/yhan94-lab/CTAC.

This study proposes an explainable few-shot learning framework for cervical cancer nest classification, addressing challenges of limited samples and dense cellular overlap. Integrating Model-Agnostic Meta-Learning (MAML) with a dual-attention feature extraction module (spatial-focused CBAM and frequency-domain FECAM), our method achieves efficient pathological feature representation in Euclidean space. Evaluated on 3,931 cervical cytopathology images (2,000 normal; 1,931 abnormal across four subtypes), the framework attained 92.3 %, $\mathbf{8 7. 5 0 \%, ~} \mathbf{9 7. 8 \%}$, and $\mathbf{9 1. 2 \%}$ on average in the normal category for Precision, Recall, Specificity, F1-Score, and Accuracy respectively. The average accuracy and variance of the four replicate experiments is $\mathbf{0. 6 9 7}$ and $6.02 \mathrm{e}-5$, respectively. It significantly outperformed 18 benchmark models. Meanwhile, the mathematical formalization validates model interpretability, while the dualattention mechanism enhances discriminative feature capture of nuclear abnormalities.

Accurate segmentation of key anatomical structures in abdominal CT images is essential for the early diagnosis and severity assessment of Metabolic Syndrome (MS). To address challenges such as blurred tissue boundaries, variable anatomical manifestations, and low contrast in medical images, we propose Met-Diff, a novel multi-organ segmentation framework based on Denoising Diffusion Probabilistic Models (DDPMs). This framework integrates several core components, including the Spatial and Channel Synergistic Attention (SCSA) module, Haar Wavelet Downsampling module, and Pre-Fusion module. These components enhance spatial-semantic feature capture, preserve detailed information during downsampling, and optimize cross-layer feature fusion respectively. By leveraging the iterative denoising mechanism inherent to diffusion models, our framework effectively captures fine morphological details and precise boundary information of target structures. Experimental results demonstrate that Met-Diff outperforms mainstream segmentation models by achieving high average Dice similarity coefficients and Intersection over Union (IoU) scores on multi-organ segmentation tasks. This work provides a reliable technical solution for MS-related medical image analysis and lays a solid foundation for computer-aided clinical diagnosis of MS.

Background and objectiveCOVID-19 is considered as the biggest global health disaster in the 21st century, and it has a huge impact on the world.MethodsThis paper publishes a publicly available dataset of CT images of multiple types of pneumonia (COVID-19CT+). Specifically, the dataset contains 409,619 CT images of 1333 patients, with subset-A containing 312 community-acquired pneumonia cases and subset-B containing 1021 COVID-19 cases. In order to demonstrate that there are differences in the methods used to classify COVID-19CT+ images across time, we selected 13 classical machine learning classifiers and 5 deep learning classifiers to test the image classification task.ResultsIn this study, two sets of experiments are conducted using traditional machine learning and deep learning methods, the first set of experiments is the classification of COVID-19 in Subset-B versus COVID-19 white lung disease, and the second set of experiments is the classification of community-acquired pneumonia in Subset-A versus COVID-19 in Subset-B, demonstrating that the different periods of the methods differed on COVID-19CT+. On the first set of experiments, the accuracy of traditional machine learning reaches a maximum of 97.3% and a minimum of only 62.6%. Deep learning algorithms reaches a maximum of 97.9% and a minimum of 85.7%. On the second set of experiments, traditional machine learning reaches a high of 94.6% accuracy and a low of 56.8%. The deep learning algorithm reaches a high of 91.9% and a low of 86.3%.ConclusionsThe COVID-19CT+ in this study covers a large number of CT images of patients with COVID-19 and community-acquired pneumonia and is one of the largest datasets available. We expect that this dataset will attract more researchers to participate in exploring new automated diagnostic algorithms to contribute to the improvement of the diagnostic accuracy and efficiency of COVID-19.

ObjectiveMelasma is a common acquired facial hyperpigmentation disorder characterized by symmetrical brown patches, often occurring in the zygomatic region, forehead, and upper lip. Its blurred boundaries, color similarity to normal skin, and irregular morphology—combined with lighting variability and skin reflections—pose significant challenges for automated lesion segmentation. This study aims to develop an effective and lightweight deep learning model tailored for accurate melasma segmentation.MethodsWe propose a novel lightweight segmentation network, HHBSNet, specifically designed for melasma lesion analysis. The model incorporates a Global Channel-Spatial Attention (GCSA) module that jointly leverages channel and spatial attention to suppress lighting interference and enhance feature discrimination in low-contrast, irregular boundaries. In addition, a Multiscale Cavity Fusion (MCF) module is introduced to extend the receptive field via multi-dilation rates, enabling effective capture of lesions at various scales without reducing resolution. The network further integrates local-global semantic fusion and adopts a combined loss strategy of cross-entropy and focal loss to address class imbalance.ResultsHHBSNet was evaluated on a self-constructed dataset comprising 501 practical facial melasma images. Quantitative results demonstrate that HHBSNet outperforms existing mainstream segmentation methods, achieving a mean Intersection over Union (Miou) of 79.69%, accuracy (ACC) of 96.68%, F-score of 88.10%, recall of 88.18%, and precision of 87.80%.ConclusionThe proposed HHBSNet demonstrates superior segmentation performance and robustness in handling melasma’s challenging visual characteristics. Its lightweight structure and strong generalization ability suggest promising potential for application in computer-aided diagnosis and large-scale clinical screening of facial pigmentary disorders.

Driver errors are the predominant factor in road accidents. Despite efforts to standardize testing conditions, external factors inherent to real-world driving environments inherently contribute to the challenge of precise driver classification. In this study, we investigate the feasibility of classifying drivers based on experience level using physiological signals collected via smart glasses equipped with electrooculography (EOG) and inertial sensors. Our methodology involved recording real-time eye and head movement data from 30 participants — 20 experienced drivers and 10 novice drivers — while navigating a predefined 28.7 km urban and highway route under natural traffic conditions. A comprehensive signal processing pipeline was developed, including median filtering, normalization, feature extraction, and statistical analysis using ANOVA and Scheffe’s method. A deep neural network classifier was then trained on the selected features, achieving an average classification accuracy of 94% across five folds. These findings demonstrate the potential of wearable sensor technologies combined with machine learning to support intelligent, non-invasive driver monitoring systems, offering personalized feedback and improving road safety.

Single-field-of-view (FoV) histopathological image classification is vital for colorectal cancer (CRC) diagnosis in mid- to low-tier hospitals lacking whole-slide imaging (WSI) scanners and storage, yet suffers from severe class imbalance and degraded performance. To address this, we propose a dual-level imbalance mitigation (DIM) framework integrating data-level and algorithm-level approaches. Specifically: (1) A global context generative adversarial network (GCGAN) generates realistic minority-class images for augmentation to balance the dataset. (2) A frequency-aware adaptive focal loss (FAFL) applies a frequency-aware offset and adaptive modulation to better separate overlapping classes. (3) A lightweight receptive field-based convolutional neural network (LRF-CNN) is trained under DIM to leverage both augmentation and loss modulation for improved classification. Extensive experiments on the single-FoV colorectal histopathology dataset demonstrate that DIM-equipped LRF-CNN outperforms five state-of-the-art models (SOTA) across multiple metrics. Furthermore, each DIM component enhances performance when applied individually to those SOTA models, and additional validation on six single-FoV histopathological datasets confirms the generalizability and effectiveness of the proposed DIM framework. Our code is available at https://github.com/Lingling-Yuan/DIM.