Journal of Imaging

Large-scale models trained on extensive datasets have become the standard due to their strong generalizability across diverse tasks. In-context learning (ICL), widely used in natural language processing, leverages these models by providing task-specific prompts without modifying their parameters. This paradigm is increasingly being adapted for computer vision, where models receive an input–output image pair, known as an in-context pair, alongside a query image to illustrate the desired output. However, the success of visual ICL largely hinges on the quality of these prompts. To address this, we propose Enhanced Instruct Me More (E-InMeMo), a novel approach that incorporates learnable perturbations into in-context pairs to optimize prompting. Through extensive experiments on standard vision tasks, E-InMeMo demonstrates superior performance over existing state-of-the-art methods. Notably, it improves mIoU scores by 7.99 for foreground segmentation and by 17.04 for single object detection when compared to the baseline without learnable prompts. These results highlight E-InMeMo as a lightweight yet effective strategy for enhancing visual ICL. Full article

Deepfake detection has become a critical issue due to the rise of synthetic media and its potential for misuse. In this paper, we propose a novel approach to deepfake detection by combining video frame analysis with facial microexpression features. The dual-branch fusion model utilizes a 3D ResNet18 for spatiotemporal feature extraction and a transformer model to capture microexpression patterns, which are difficult to replicate in manipulated content. We evaluate the model on the widely used FaceForensics++ (FF++) dataset and demonstrate that our approach outperforms existing state-of-the-art methods, achieving 99.81% accuracy and a perfect ROC-AUC score of 100%. The proposed method highlights the importance of integrating diverse data sources for deepfake detection, addressing some of the current limitations of existing systems. Full article

Microscopic image automatic recognition is a core technology for mineral composition analysis and plays a crucial role in advancing the intelligent development of smart mining systems. To overcome the limitations of traditional lithium ore analysis and meet the challenges of deployment on edge devices, we propose PS-YOLO-seg, a lightweight segmentation model specifically designed for lithium mineral analysis under visible light microscopy. The network is compressed by adjusting the width factor to reduce global channel redundancy. A PSConv-based downsampling strategy enhances the network’s ability to capture dim mineral textures under microscopic conditions. In addition, the improved C3k2-PS module strengthens feature extraction, while the streamlined Segment-Efficient head minimizes redundant computation, further reducing the overall model complexity. PS-YOLO-seg achieves a slightly improved segmentation performance compared to the baseline YOLOv12n model on a self-constructed lithium ore microscopic dataset, while reducing FLOPs by 20%, parameter count by 33%, and model size by 32%. Additionally, it achieves a faster inference speed, highlighting its potential for practical deployment. This work demonstrates how architectural optimization and targeted enhancements can significantly improve instance segmentation performance while maintaining speed and compactness, offering strong potential for real-time deployment in industrial settings and edge computing scenarios. Full article

Ultrasound attenuation imaging (ATI) is a non-invasive method for quantifying hepatic steatosis, offering advantages over the hepatorenal index (HRI). However, its reliability can be influenced by factors such as measurement depth, ROI size, and subcutaneous fat. This paper examines the impact of these confounders on ATI measurements and discusses diagnostic considerations. In this study, 33 healthy adults underwent liver ultrasound with ATI and HRI protocols. ATI measurements were taken at depths of 2–5 cm below the liver capsule using small and large ROIs. Two operators performed the measurements, and inter-operator variability was assessed. Subcutaneous fat thickness was measured to evaluate its influence on attenuation values. The ATI measurements showed a consistent decrease in attenuation coefficient values with increasing depth, approximately 0.05 dB/cm/MHz. Larger ROI sizes increased measurement variability due to greater anatomical heterogeneity. HRI values correlated weakly with ATI and were influenced by operator technique and subcutaneous fat, the latter accounting for roughly 2.5% of variability. ATI provides a quantitative assessment of hepatic steatosis compared to HRI, although its accuracy can vary depending on the depth and ROI selection. Standardised imaging protocols and AI tools may improve reproducibility and clinical utility, supporting advancements in ultrasound-based liver diagnostics for better patient care. Full article

In this paper, we propose an innovative essential multi-secret image sharing (EMSIS) scheme that integrates sensor data to securely and efficiently share multiple secret images of varying importance. Secret images are categorized into hierarchical levels and encoded into essential shadows and fault-tolerant non-essential shares, with access to higher-level secrets requiring higher-level essential shadows. By incorporating sensor data, such as location, time, or biometric input, into the encoding and access process, the scheme enables the context-aware and adaptive reconstruction of secrets based on real-world conditions. Experimental results demonstrate that the proposed method not only strengthens hierarchical access control, but also enhances robustness, flexibility, and situational awareness in secure image distribution systems. Full article

AI-generated content has grown significantly in recent years. Today, AI-generated and human-made images coexist across various settings, including news media, social platforms, and beyond. However, we still know relatively little about how audiences interpret and evaluate these different types of images. The goal of this study was to examine whether image interpretation is influenced by the origin of the image (AI-generated vs. human-made). Additionally, we aimed to explore whether visual professionalization influences how images are interpreted. To this end, we presented 24 AI-generated images (produced using Midjourney, DALL·E, and Firefly) and 8 human-made images to 161 participants—71 visual professionals and 90 non-professionals. Participants were asked to evaluate each image based on the following: (1) the source they believed the image originated from, (2) the level of realism, and (3) the level of credibility they attributed to it. A total of 5152 responses were collected for each question. Our results reveal that human-made images are more readily recognized as such, whereas AI-generated images are frequently misclassified as human-made. We also find that human-made images are perceived as both more realistic and more credible than AI-generated ones. We conclude that individuals are generally unable to accurately determine the source of an image, which in turn affects their assessment of its credibility. Full article

Traditionally, Helicobacter pylori (HP) gastritis has been diagnosed by pathologists through the examination of gastric biopsies using optical microscopy with standard hematoxylin and eosin (H&E) staining. However, with the adoption of digital pathology, the identification of HP faces certain limitations, particularly due to insufficient resolution in some scanned images. Moreover, interobserver variability has been well documented in the traditional diagnostic approach, which may further complicate consistent interpretation. In this context, deep convolutional neural network (DCNN) models are showing promising results in the automated detection of this infection in whole-slide images (WSIs). The aim of the present article is to detect the presence of HP infection from our own institutional dataset of histopathological gastric biopsy samples using different pretrained and recognized DCNN and AutoML approaches. The dataset comprises 100 H&E-stained WSIs of gastric biopsies. HP infection was confirmed previously using immunohistochemical confirmation. A total of 45,795 patches were selected for model development. InceptionV3, Resnet50, and VGG16 achieved AUC (area under the curve) values of 1. However, InceptionV3 showed superior metrics such as accuracy (97%), recall (100%), F1 score (97%), and MCC (93%). BoostedNet and AutoKeras achieved accuracy, precision, recall, specificity, and F1 scores less than 85%. The InceptionV3 model was used for external validation, and the predictions across all patches yielded a global accuracy of 78%. In conclusion, DCNN models showed stronger potential for diagnosing HP in gastric biopsies compared with the auto ML approach. However, due to variability across pathology applications, no single model is universally optimal. A problem-specific approach is essential. With growing WSI adoption, DL can improve diagnostic accuracy, reduce variability, and streamline pathology workflows using automation. Full article

Color vision deficiency (CVD) affects around 300 million people globally due to issues with cone cells, highlighting the need for effective daltonization methods. These methods modify color palettes to enhance detail visibility for individuals with CVD. However, they can also distort the natural appearance of images. This study presents a novel daltonization method that focuses on preserving image naturalness for both normal trichromats and individuals with CVD. Our approach modifies only the achromatic component while enhancing detail visibility for individuals with CVD. To compare our approach with the previously known anisotropic daltonization method, we utilize objective and subjective evaluations that separately assess visibility enhancement and naturalness preservation. Our findings indicate that the proposed method outperforms the anisotropic method in naturalness by over 10 times according to objective criteria. Subjective evaluations revealed that more than 90% of CVD individuals and 95% of trichromats preferred our method for its natural appearance. Although objective contrast metrics suggest inferior visibility enhancement, subjective evaluation indicates comparable performance: contrast improvement was observed in 65% of protan cases for our method versus 70% for the anisotropic method, with contrast deterioration in 18% versus 7%, respectively. Overall, our method offers superior naturalness while maintaining comparable detail discrimination. Full article

Sex determination is a fundamental component of forensic identification and medicolegal investigations. Several studies have investigated sexual dimorphism through mandibular osteometric measurements, including the position of anatomical foramina such as the mandibular and mental foramen (MF), reporting population-specific discrepancies. This study assessed the reliability and predictive ability of specific anthropometric mandibular measurements for sex estimation using three-dimensional (3D) cone beam computed tomography (CBCT) surface reconstructions. Methods: CBCT scans from 204 Greek individuals (18–70 years) were analyzed. Records were categorized by sex and age. Five linear measurements were traced on 3D reconstructions using ViewBox 4 software: projections of the inferior points of the right and left mental and mandibular foramina and the linear distance between mental foramina projections. A binary logistic regression (BLR) model was employed. All measurements showed statistically significant sex differences, with males presenting higher mean values. The final model achieved accuracy of 66.7% in sex prediction, with two vertical measurements—distances from the right mandibular foramen and the left mental foramen—identified as the strongest predictors of sex. The positions of the mandibular and mental foramina demonstrate sex-related dimorphism in this Greek sample, supporting their forensic relevance in population-specific applications. Full article

Transverse fetal head position during labor is associated with increased rates of operative deliveries and cesarean sections. Traditional assessment methods rely on digital examination, which can be inaccurate in cases of prolonged labor. Intrapartum ultrasound offers improved diagnostic capabilities, but standardized interpretation frameworks are needed. This study aimed to evaluate the significance of appropriate assessment and management of transverse fetal head position during labor, with particular emphasis on the correlation between geometric parameters and delivery outcomes. Additionally, the investigation analyzed the potential role of Artificial Intelligence Dystocia Algorithm (AIDA) as an innovative decision support system in standardizing diagnostic approaches and optimizing clinical decision-making in cases of fetal malposition. This investigation was conducted as a focused secondary analysis of data originally collected for the development and validation of the Artificial Intelligence Dystocia Algorithm (AIDA). The study examined 66 cases of transverse fetal head position from a cohort of 135 nulliparous women with prolonged second-stage labor across three Italian hospitals. Cases were stratified by Midline Angle (MLA) measurements into classic transverse (≥75°), near-transverse (70–74°), and transitional (60–69°) positions. Four geometric parameters (Angle of Progression, Head–Symphysis Distance, Midline Angle, and Asynclitism Degree) were evaluated using the AIDA classification system. The predictive capabilities of three machine learning algorithms (Support Vector Machine, Random Forest, and Multilayer Perceptron) were assessed, and delivery outcomes were analyzed. The AIDA system successfully categorized labor dystocia into five distinct classes, with strong predictive value for delivery outcomes. A clear gradient of cesarean delivery risk was observed across the spectrum of transverse positions (100%, 93.1%, and 85.7% for near-transverse, classic transverse, and transitional positions, respectively). All cases classified as AIDA Class 4 required cesarean delivery regardless of the specific MLA value. Machine learning algorithms demonstrated high predictive accuracy, with Random Forest achieving 95.5% overall accuracy across the study cohort. The presence of concurrent asynclitism with transverse position was associated with particularly high rates of cesarean delivery. Among the seven cases that achieved vaginal delivery despite transverse positioning, none belonged to the classic transverse positions group, and five (71.4%) exhibited at least one parameter classified as favorable. The integration of artificial intelligence through AIDA as a decision support system, combined with intrapartum ultrasound, offered a promising approach for objective assessment and management of transverse fetal head position. The AIDA classification system’s integration of multiple geometric parameters, with particular emphasis on precise Midline Angle (MLA) measurement in degrees, provided superior predictive capability for delivery outcomes compared to qualitative position assessment alone. This multidimensional approach enabled more personalized and evidence-based management of malpositions during labor, potentially reducing unnecessary interventions while identifying cases where expectant management might be futile. Further prospective studies are needed to validate the predictive capability of this decision support system and its impact on clinical decision-making in real-time labor management. Full article

Thoracic injuries account for a high percentage of combat casualty mortalities, with 80% of preventable deaths resulting from abdominal or thoracic hemorrhage. An effective method for detecting and triaging thoracic injuries is point-of-care ultrasound (POCUS), as it is a cheap and portable noninvasive imaging method. POCUS image interpretation of pneumothorax (PTX) or hemothorax (HTX) injuries requires a skilled radiologist, which will likely not be available in austere situations where injury detection and triage are most critical. With the recent growth in artificial intelligence (AI) for healthcare, the hypothesis for this study is that deep learning (DL) models for classifying images as showing HTX or PTX injury, or being negative for injury can be developed for lowering the skill threshold for POCUS diagnostics on the future battlefield. Three-class deep learning classification AI models were developed using a motion-mode ultrasound dataset captured in animal study experiments from more than 25 swine subjects. Cluster analysis was used to define the “population” based on brightness, contrast, and kurtosis properties. A MobileNetV3 DL model architecture was tuned across a variety of hyperparameters, with the results ultimately being evaluated using images captured in real-time. Different hyperparameter configurations were blind-tested, resulting in models trained on filtered data having a real-time accuracy from 89 to 96%, as opposed to 78–95% when trained without filtering and optimization. The best model achieved a blind accuracy of 85% when inferencing on data collected in real-time, surpassing previous YOLOv8 models by 17%. AI models can be developed that are suitable for high performance in real-time for thoracic injury determination and are suitable for potentially addressing challenges with responding to emergency casualty situations and reducing the skill threshold for using and interpreting POCUS. Full article

This meta-analysis evaluated the diagnostic accuracy of sonoelastography for distinguishing benign and malignant breast lesions, comparing strain elastography and shear wave elastography (SWE). We systematically reviewed 825 publications, selecting 30 studies (6200 lesions: 45% benign, 55% malignant). The pooled sensitivity and specificity for overall sonoelastography were 88% (95% CI: 85–91%) and 84% (95% CI: 81–87%), respectively. Strain elastography showed sensitivity and specificity of 85% and 80%, respectively, while SWE demonstrated superior performance with 90% sensitivity, 86% specificity, and an AUC of 0.92. Moderate heterogeneity (I² = 55%) was attributed to study variation. SWE showed the potential to reduce unnecessary biopsies by 30–40% by increasing specificity. AI-assisted image analysis and standardized protocols may enhance accuracy and reduce variability. These findings support the integration of SWE into breast imaging protocols. Full article

This study presents a novel framework that integrates Vision Graph Neural Networks (ViGs) with supervised contrastive learning for enhanced spectro-temporal image analysis of speech signals in Parkinson’s disease (PD) detection. The approach introduces a frequency band decomposition strategy that transforms raw audio into three complementary spectral representations, capturing distinct PD-specific characteristics across low-frequency (0–2 kHz), mid-frequency (2–6 kHz), and high-frequency (6 kHz+) bands. The framework processes mel multi-band spectro-temporal representations through a ViG architecture that models complex graph-based relationships between spectral and temporal components, trained using a supervised contrastive objective that learns discriminative representations distinguishing PD-affected from healthy speech patterns. Comprehensive experimental validation on multi-institutional datasets from Italy, Colombia, and Spain demonstrates that the proposed ViG-contrastive framework achieves superior classification performance, with the ViG-M-GELU architecture achieving 91.78% test accuracy. The integration of graph neural networks with contrastive learning enables effective learning from limited labeled data while capturing complex spectro-temporal relationships that traditional Convolution Neural Network (CNN) approaches miss, representing a promising direction for developing more accurate and clinically viable speech-based diagnostic tools for PD. Full article

Microscopic cell classification is a fundamental challenge in both clinical diagnosis and biological research. However, existing methods still struggle with the complexity and morphological diversity of cellular images, leading to limited accuracy or high computational costs. To overcome these constraints, we propose an efficient classification method that balances strong feature representation with a lightweight design. Specifically, an Inception-Linear Attention-based Lightweight Vision Transformer (ILViT) model is developed for microscopic cell classification. The ILViT integrates two innovative modules: Dynamic Inception Convolution (DIC) and Contrastive Omni-Kolmogorov Attention (COKA). DIC combines dynamic and Inception-style convolutions to replace large kernels with fewer parameters. COKA integrates Omni-Dimensional Dynamic Convolution (ODC), linear attention, and a Kolmogorov-Arnold Network(KAN) structure to enhance feature learning and model interpretability. With only 1.91 GFLOPs and 8.98 million parameters, ILViT achieves high efficiency. Extensive experiments on four public datasets are conducted to validate the effectiveness of the proposed method. It achieves an accuracy of 97.185% on BioMediTech dataset for classifying retinal pigment epithelial cells, 97.436% on ICPR-HEp-2 dataset for diagnosing autoimmune disorders via HEp-2 cell classification, 90.528% on Hematological Malignancy Bone Marrow Cytology Expert Annotation dataset for categorizing bone marrow cells, and 99.758% on a white blood cell dataset for distinguishing leukocyte subtypes. These results show that ILViT outperforms the state-of-the-art models in both accuracy and efficiency, demonstrating strong generalizability and practical potential for cell image classification. Full article

Self-supervised depth estimation from monocular image sequences provides depth information without costly sensors like LiDAR, offering significant value for autonomous driving. Although self-supervised algorithms can reduce the dependence on labeled data, the performance is still affected by scene occlusions, lighting differences, and sparse textures. Existing methods do not consider the enhancement and interaction fusion of features. In this paper, we propose a novel parallel multi-scale semantic-depth interactive fusion network. First, we adopt a multi-stage feature attention network for feature extraction, and a parallel semantic-depth interactive fusion module is introduced to refine edges. Furthermore, we also employ a metric loss based on semantic edges to take full advantage of semantic geometric information. Our network is trained and evaluated on KITTI datasets. The experimental results show that the methods achieve satisfactory performance compared to other existing methods. Full article

Soft tissue masses are predominantly benign, with a benign-to-malignant ratio exceeding 100:1, often located around joints. They may be contiguous or adjacent to joints or reflect systemic diseases or distant organ involvement. Clinically, they typically present as palpable swellings. Evaluation should consider duration, size, depth, and mobility. Also assess consistency, growth rate, symptoms, and history of trauma, infection, or malignancy. Laboratory tests are generally of limited diagnostic value. The primary clinical goal is to avoid unnecessary investigations or procedures for benign lesions while ensuring timely diagnosis and treatment of malignant ones. Imaging plays a central role: it confirms the presence of the lesion, assesses its location, size, and composition, differentiates between cystic and solid or benign and malignant features, and can sometimes provide a definitive diagnosis. Imaging is also crucial for biopsy planning, treatment strategy, identification of involved structures, and follow-up. Ultrasound (US) is the first-line imaging modality for palpable soft tissue masses due to its low cost, wide availability, and lack of ionizing radiation. If findings are inconclusive, magnetic resonance imaging (MRI) or computed tomography (CT) is recommended. This review aims to discuss the most common causes of periarticular soft tissue masses in the appendicular skeleton, focusing on clinical presentation and radiologic features. Full article

Meningiomas are the most common primary intracranial tumors in adults. For most cases, surgical resection is effective in mitigating recurrence risk. Accurate visualization of meningiomas helps radiologists assess the distribution and volume of the tumor within the brain while assisting neurosurgeons in preoperative planning. This paper introduces an innovative realistic 3D medical visualization system, namely SegR3D. It incorporates a 3D medical image segmentation pipeline, which preprocesses the data via semi-supervised learning-based multi-target segmentation to generate masks of the lesion areas. Subsequently, both the original medical images and segmentation masks are utilized as non-scalar volume data inputs into the realistic rendering pipeline. We propose a novel importance transfer function, assigning varying degrees of importance to different mask values to emphasize the areas of interest. Our rendering pipeline integrates physically based rendering with advanced illumination techniques to enhance the depiction of the structural characteristics and shapes of lesion areas. We conducted a user study involving medical practitioners to evaluate the effectiveness of SegR3D. Our experimental results indicate that SegR3D demonstrates superior efficacy in the visual analysis of meningiomas compared to conventional visualization methods. Full article

Two-dimensional imaging is still commonly used in dentistry, but does not provide the three-dimensional information often required for the accurate assessment of dental structures. Photon-counting computed tomography (PCCT), a new three-dimensional modality mainly used in general medicine, has shown promising potential for dental applications. With growing digitalization and cross-disciplinary integration, using PCCT data from other medical fields is becoming increasingly relevant. Conventional CT scans, such as those of the cervical spine, have so far lacked the resolution to reliably evaluate dental structures or implants. This study evaluates the diagnostic utility of PCCT for visualizing peri-implant structures in routine clinical photon-counting CT acquisitions and assesses the influence of metal artifact reduction (MAR) algorithms on image quality. Ten dental implants were retrospectively included in this IRB-approved study. Standard PCCT scans were reconstructed at multiple keV levels with and without MAR. Quantitative image analysis was performed with respect to contrast and image noise. Qualitative evaluation of peri-implant tissues, implant shoulder, and apex was performed independently by two experienced dental professionals using a five-point Likert scale. Inter-reader agreement was measured using intraclass correlation coefficients (ICCs). PCCT enabled high-resolution imaging of all peri-implant regions with excellent inter-reader agreement (ICC > 0.75 for all structures). Non-MAR reconstructions consistently outperformed MAR reconstructions across all evaluated regions. MAR led to reduced clarity, particularly in immediate peri-implant areas, without significant benefit from energy level adjustments. All imaging protocols were deemed diagnostically acceptable. This is the first in vivo study demonstrating the feasibility of opportunistic dental diagnostics using PCCT in a clinical setting. While MAR reduces peripheral artifacts, it adversely affects image clarity near implants. PCCT offers excellent image quality for peri-implant assessments and enables incidental detection of dental pathologies without additional radiation exposure. PCCT opens new possibilities for opportunistic, three-dimensional dental diagnostics during non-dental CT scans, potentially enabling earlier detection of clinically significant pathologies. Full article