Search Results (519)

This paper investigates human–AI collaboration through explainable AI where we interpret the results of barefoot print clustering using unsupervised machine learning. This can be used to identify the number of individuals from barefoot prints on the ground as a tool in forensics or anti-poaching. A self-supervised vision transformer, DINOv2, is used to automatically extract feature embeddings from localized barefoot-print regions to identify trails belonging to an individual on soil substrate. Furthermore, we introduce an Embedding Spatial Attribution Module (ESAM) to generate spatial attribution heatmaps, enabling visualization of discriminative regions that contribute to individual-specific trail identification and improving model explainability. The proposed method is named XTrail-ID, an explainable human footprint trail identification framework with two variants, OBB-XTrail-ID (oriented bounding box-based), and SEG-XTrail-ID (segmentation-based). We quantify embedding similarity using three complementary metrics: cosine similarity, Pearson correlation coefficient, and Spearman rank correlation. Twenty adults (ten males, ten females) participated, with a total of 1000 trail images extracted from UAV imagery. SEG-XTrail-ID using cosine similarity yielded the highest performance, with (3.21) discriminability and (94.2%) accuracy, while OBB-XTrail-ID using cosine similarity achieved (2.54) discriminability and (91.5%) accuracy. In addition, the latter exhibited reduced consistency in footprint grouping when more than three individuals were present within a single frame. Full article

Background/Objectives: Survivors of sexual violence and domestic violence/intimate partner violence (IPV) often require support beyond the immediate emergency encounter; however, post-acute care remains inconsistently defined, unevenly organised or conceptualised, and fragmented across service systems. This scoping review mapped international post-acute follow-up, care, assistance, and support pathways, with particular attention to organisational models, continuity mechanisms, loss to follow-up after first access, and implications for the Italian context. Methods: We conducted an international health-services scoping review of post-acute follow-up, care, assistance, and support interventions for survivors of sexual violence and domestic violence/IPV. Searches were performed in PubMed/MEDLINE, Scopus, Web of Science Core Collection, Embase, APA PsycINFO via EBSCOhost, and CINAHL via EBSCOhost. Eligible studies were published from 2013 onward and had to describe an identifiable post-acute component beyond the initial emergency, forensic, or first-contact phase. The review followed a Population–Concept–Context framework and was reported in accordance with PRISMA-ScR. Results: Forty-four studies were included in the core synthesis, comprising 16 studies on sexual violence/sexual assault, 27 on domestic violence/IPV, and one mixed domestic, family, and sexual violence outreach model. The sexual violence literature clustered around early trauma-focused interventions, sexual assault care centre pathways, medical follow-up, follow-up attendance, and digital continuity tools. The IPV literature was broader and included psychotherapy, advocacy and case-management models, housing-first and trauma-informed stabilisation approaches, nurse-led and clinic-based services, outreach and safety-contact programmes, digital interventions, and programmes for system-involved survivors. Across both fields, the pathways most consistently described as supporting continuity combined structured re-contact, coordinated support, and multi-component responses over time. Conclusions: The mapped literature supports conceptualising post-acute responses to sexual violence and domestic violence/IPV as continuity pathways that extend beyond first contact and link healthcare, psychological, advocacy, and social supports. Systems may be better positioned to support continuity when they provide structured follow-up, warm handoffs, coordinated navigation, and context-sensitive recovery models. These findings point to provisional, evidence-informed organisational questions for strengthening post-acute pathways, including in Italy, particularly around structured re-contact, warm handoffs, survivor navigation, and integration between healthcare, anti-violence, psychological, and territorial social-support services. Full article

Background/Objectives: Anthropometric measurements provide essential normative datasets that form the foundation for clinical practice and forensic identification. The human ear is a highly informative structure due to its complex morphology and individual specificity, making it a valuable tool for biometric systems. This study aimed to estimate biological sex based on auricular morphometric measurements, develop a logistic regression model for this purpose, and validate its performance using ROC analysis. Materials and Methods: This cross-sectional study included 120 adult participants (60 males, 60 females). Standardized digital photographs were analyzed in ImageJ to record 22 linear and 6 angular measurements using established anatomical landmarks. LASSO logistic regression was employed for variable selection and model shrinkage. The final model’s discriminative performance was assessed using the area under the receiver operating characteristic (ROC) curve (AUC), the Hosmer–Lemeshow test, and the Brier score. Results: A comparative analysis revealed that most linear and angular measurements showed significant sexual dimorphism. Almost all linear dimensions (A1–A22) were significantly larger in males (p < 0.001). Auricular width (A2) and width at the level of the tragus (A3) emerged as the most robust indicators, demonstrating “very large” effect sizes. Conversely, the angle between the preauricular line and the vertical plane (A28) was significantly greater in females, providing a unique inverse relationship for sex estimation. A parsimonious 5-predictor model (incorporating A2, A3, A5, A10, and A28) achieved exceptional discriminative performance with an AUC of 0.980. Conclusions: Auricular morphometry is a highly effective tool for sex estimation. The findings confirm significant sexual dimorphism in the external ear, particularly in linear dimensions. The developed model may serve as a preliminary morphometric reference for future automated biometric recognition studies, although no artificial intelligence-based classification model was developed in the present study. Full article

Background/Objectives: Determining chronological age is important in several domains, including forensic identification, clinical decision-making, legal matters, and immigration procedures. Dental tissues are widely recognized as reliable indicators of age because they undergo gradual and measurable structural changes throughout life. Nevertheless, most conventional dental methods show limited reliability when applied to adults and elderly individuals. The objective of this study was to investigate an automated deep learning-based approach for age-group classification in adults and seniors using panoramic dental radiographs. Methods: Panoramic dental radiographs were analyzed using a custom-designed Convolutional Neural Network (CNN) along with several established pre-trained deep learning architectures. The dataset consisted of 1469 radiographic images obtained from Egyptian individuals aged between 25 and 70 years. Images were classified into five predefined age categories using a classification-based framework, and the models were trained to learn age-related dental patterns from radiographic images. Results: The proposed Custom CNN achieved the highest accuracy of 85.2%, outperforming YOLOv8 (79.1%) and all other evaluated models, with the lowest prediction error (MAE = 1.92 years; RMSE = 5.46 years). Overall, the deep learning models demonstrated strong performance in classifying dental age groups, particularly within adult and senior populations, where conventional methods often show reduced reliability. Conclusions: The findings suggest that deep learning analysis of panoramic dental radiographs may serve as a supportive tool for age-group classification in adult populations, complementing rather than replacing traditional assessment methods. These results, while promising, are limited to the dataset and experimental conditions of this study, and broader applicability requires further validation across diverse populations and settings. Full article

Background/Objectives: Accurate bone lesion classification on radiographs is critical for clinical decision-making and forensic identification. Existing deep learning approaches treat radiographs as whole images, neglecting available spatial annotations and clinical metadata. To develop an ROI-guided deep learning framework integrating clinical metadata for three-class (Normal, Benign, Malignant) bone lesion classification and to assess its clinical safety profile. Methods: Using the BTXRD (3746 radiographs: 1879 Normal, 1525 Benign, 342 Malignant), an EfficientNetV2-S backbone was combined with an 11-dimensional metadata MLP trained on ROI-cropped regions. Training employed Focal Loss with adaptive class weighting, Mixup/CutMix augmentations, Stochastic Weight Averaging, and Test-Time Augmentation. Five-fold stratified cross-validation with bootstrap confidence intervals (n = 2000) and probability calibration metrics were used. Results: The framework achieved 96.05% accuracy (95% CI: 95.41–96.66%), 93.94% balanced accuracy, 92.62% macro F1-score, and 99.21% macro-AUC (95% CI: 98.89–99.42%). Critically, near-zero Malignant-to-Normal misclassifications occurred (1/342, 0.29%; 95% Clopper–Pearson CI: 0.01–1.62%) across all 3746 predictions. The minority Malignant class attained F1 = 83.53% despite comprising only 9.1% of the dataset. Conclusions: ROI-guided deep learning with metadata fusion achieves state-of-the-art bone lesion classification with clinically safe error patterns and probability outputs whose calibration was explicitly quantified, supporting its potential as a decision support tool in diagnostic radiology and forensic anthropology, pending external validation on independent cohorts. Full article

The expanding capabilities of deep learning-based media synthesis have intensified concerns regarding the authenticity of digital content and the reliability of forensic analysis tools. In response to these challenges, this work introduces DeepFakeX, a collection of 800 synthetically generated videos available under controlled access for research purposes. The dataset encompasses four distinct categories of AI-driven synthesis: facial identity replacement, audio track substitution, neural voice cloning, and combined audiovisual alteration. Unlike existing deepfake datasets that predominantly focus on facial synthesis, DeepFakeX covers a broader range of manipulation modalities, reflecting the diversity of synthetic media encountered in real-world settings. All deepfakes were generated using state-of-the-art, publicly available tools. Standardized post-processing procedures were applied to each video to ensure uniformity in terms of quality, duration and encoding format. DeepFakeX also emphasizes diversity in gender, age, ethnicity, and language. Video contexts span speeches, informational videos, movie clips, news broadcasts, and interviews that reflect content scenarios commonly encountered in real-world online environments. The dataset includes videos in both English and Urdu. The dataset’s quality and structural variability were assessed through visual and audio analyses using the Structural Similarity Index Measure (SSIM), Mel-Frequency Cepstral Coefficients (MFCCs), and Principal Component Analysis (PCA). The evaluation results revealed substantial variability within each manipulation category, along with clearly distinguishable patterns specific to each modality. DeepFakeX has been developed to facilitate rigorous and transparent research in deepfake detection, cross-modal forensic analysis, and AI-driven media forensics. It is hosted on Zenodo under controlled access for research use. Full article

Automatic dental age estimation in deep learning applications relies on accurate assessment of tooth development stage (TDS) and regions of interest (ROIs). Since dental development is population-specific, a recent study showed that a two-step approach using an ethnicity-specific reference dataset can improve explainability, transferability across populations, and agreement with chronological age. Therefore, this study aims to investigate whether the proposed population-aware framework can be extended toward greater automation by using Faster Region-Based Convolutional Network (R-CNN) for third-molar detection. A total of 1110 digital panoramic radiographs and 1980 labeled ROIs were used to train and validate four pretrained backbones: AlexNet, DenseNet-201, GoogLeNet, and VGG-16. DenseNet-201 achieved the best training performance and overall validation performance (accuracy = 95% and marco recall = 0.75), followed by GoogLeNet (84.7% and 0.80), VGG-16 (86.6% and 0.75), and AlexNet (79.8% and 0.58). Validation results showed relatively high recall across most TDS stages, whereas precision was lower because of false-positive (FP) detections. Performance was only minimally affected by the intersection-over-union (IoU) threshold, suggesting stable localization behavior. The trained Faster R-CNN models provide a clinical-oriented assistive tool for automatic third-molar detection and TDS classification, serving as an intermediate component that can potentially support more efficient ethnicity-specific dental age estimation in future clinical practice, forensic investigations, and pediatric dentistry. Full article

Digital evidence management systems are designed to ensure that the digital evidence is genuine and effectively handle its complexity. In this work, blockchain technology is applied to handle the digital evidence by introducing several layers of security to ensure its protection, data integrity, and confidentiality, as well as trace the evidence throughout all its phases. To store the evidence files and their metadata, the proposed system uses a decentralized storage architecture that utilizes the InterPlanetary File System (IPFS) and Google Drive. Moreover, the proposed system ensures the chain of custody of the digital evidence through the use of Hyperledger Fabric technology. In addition, smart contracts (chaincode) are used in this work to validate the digital evidence, enforce strong access controls, and protect evidence metadata integrity. To ensure reliable transaction sequencing and consistency across the distributed ledger, an ordering service is used. At last, we combine two hash algorithms, symmetric encryption, file fragmentation, and metadata logging to protect the digital evidence from unauthorized access. The proposed framework is integrated with modern forensic tools, including Autopsy. The procedure of acquiring and analyzing digital evidence is made straightforward by the application of a set of forensic procedures. Moreover, the system’s modular design allows users to perform preprocessing operations, administer the decentralized storage, administer the evidence retrieval, test system performance, and enhance the system scalability. Moreover, we implemented secure coding practices and applied large language models to mitigate identified vulnerabilities, including weak system input validation, concurrent access to the system, and an insecure logging system. The experimental results indicate that the proposed framework preserves the digital evidence’s integrity, ensures chain of custody, and records all transactions. Results also indicate that the digital evidence is protected from unauthorized access and change attempts. Finally, by following local relevant regulations and established standards, the digital evidence should be admissible in court. Full article

Digital evidence (DE) continues to evolve alongside rapid technological innovation, including the increasing integration of AI-generated media, algorithmic tools, and digital forensic technologies used to identify, extract, and analyse digital artefacts in criminal investigations. Yet, limited work has examined how frontline police officers interpret and operationalise digital forensic outputs, particularly in the context of emerging forms of AI-mediated evidence. We investigated the perceptions of 13 police officers in England and Wales regarding the evolving role of DE and digital forensic investigations, and the implications for policing practice. Reflexive thematic analysis identified four themes: (i) AI and the evolving landscape of digital forensic investigations; (ii) digital systems and the infrastructure of investigations; (iii) human judgement and trust in the interpretation of DE; and (iv) building digital expertise in modern investigations. Participants described a marked rise in highly realistic AI-generated imagery, which complicates evidence categorisation and proportionality assessments. This also reshapes investigative decision-making while necessitating adaptations in investigative interviewing, particularly around disclosure sequencing and evidential challenge. Findings suggest that understanding of AI and algorithmic systems among some frontline officers remains underdeveloped, raising concerns about the interpretation of outputs. By foregrounding practitioner perspectives, this study contributes a human-centric understanding of digital forensic practice, while offering insights into the future development of investigative approaches in response to emerging technologies and evolving threats. Full article

The growing fragmentation of digital evidence in modern computing environments poses significant challenges for digital forensic analysis. Data is often deleted, overwritten, or distributed across heterogeneous platforms, limiting the effectiveness of traditional forensic tools that rely on intact files and deterministic rules. This work addresses a key limitation in current forensic methodologies: the scarcity of learning-based approaches capable of identifying patterns in fragmented and incomplete digital evidence. To address this challenge, we propose PatternMiner, a hybrid deep learning framework that integrates Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and Transformer encoders. The framework combines byte-level content fragments with contextual metadata, such as timestamps and file permissions, enabling multimodal inference from fragmented data while explicitly excluding label-derived features to prevent leakage. PatternMiner is evaluated on established forensic benchmark datasets, including Digital Corpora and AFF4 forensic containers, which simulate realistic fragmentation scenarios. All experiments are conducted under an explicit leakage-controlled evaluation protocol with group-aware data partitioning to ensure that performance reflects generalization to unseen data. Results show that the proposed framework achieves strong performance, with an accuracy of 92.1% and a macro-averaged F1-score of 92.1% under complete input conditions. Furthermore, the model demonstrates resilience to degraded and partially corrupted inputs, including truncation, byte removal, shifting, and fragment reordering. These findings indicate that PatternMiner effectively captures structural and contextual patterns in fragmented data, providing a practical step toward more reliable and data-driven forensic analysis. By combining multimodal learning with rigorous evaluation practices, the proposed framework contributes to developing scalable and generalizable solutions for modern digital forensic environments. Full article

Introduction: The global spread of the SARS-CoV-2 pandemic led to a serious health, social and economic global crisis. This pandemic was and remains the most important health emergency worldwide, for which all professionals have been called to provide diagnosis and treatment support. Despite early concerns about safety, forensic medicine has contributed to a better understanding of the pathological mechanisms involved. Objective: This study aims to describe and analyze the postmortem pathological findings in confirmed SARS-CoV-2 cases, emphasizing the contribution of forensic autopsies to elucidating the mechanisms of death and associated comorbidities. Methods: A retrospective study was conducted on 279 autopsies between 2020 and 2022. Demographic, clinical, and pathological data were collected and statistically analyzed. Results: Following the descriptive analysis of the cases included in the study, as well as the analysis of the relevant scientific literature, the major impact of the COVID-19 pandemic was highlighted in terms of the death mechanisms involved, occurring consequences and induced changes. Conclusions: Autopsies remain the essential tools for investigating COVID-19-related deaths. The findings confirm that SARS-CoV-2 primarily affects the pulmonary and cardiovascular systems. However, the overlap between “death from” and “death with” COVID-19 highlights the need for standardized postmortem diagnostic criteria and comprehensive clinical–pathological correlation. Full article

Genome-edited (GE) crops are reaching consumers faster than the analytical infrastructure designed to monitor them. Unlike transgenic crops, most GE products carry only small sequence changes and no foreign DNA, making conventional element-based polymerase chain reaction (PCR) screening, which has underpinned genetically modified organism (GMO) enforcement for two decades, largely ineffective. This review critically evaluates detection technologies not by listing them sequentially, but by comparing their performance against a shared set of enforcement-relevant criteria: sensitivity at regulatory thresholds, allele discrimination capacity, prior target knowledge requirement, and validation maturity. Building on detection/discrimination distinctions already present in ENGL guidance documents and the DETECT project, we formalize a two-axis framework separating detectability (technically achievable for most known targets in defined seed or DNA mixtures at or near the 0.1% MRPL) from identifiability (rarely achievable without developer disclosure), with detection as a necessary precondition for identification, and apply it product by product to each commercialized GE crop for which public molecular data are available. The sulfonylurea-tolerant canola (SU Canola) case, in which analytical specificity is established but forensic event specificity is contested, and the German DETECT project are examined as contrasting case studies of analytical success and attribution failure, extracting generalizable lessons for the field. A technology comparison table, a product-specific feasibility matrix, and a tiered enforcement workflow are provided as practical tools. We conclude with five research priorities for closing the detection–identification gap across near-term, mid-term and longer-term horizons. Full article

Mitochondrial DNA (mtDNA) analysis is a fundamental tool in forensic genetics, particularly when biological samples exhibit severe degradation or low nuclear DNA content. Its unique biological characteristics, such as a high copy number per cell, strict matrilineal inheritance, and lack of recombination, enable human identification and reconstruction of maternal lineages in complex contexts, including disaster victim identification, historical cases, and missing persons investigations. This narrative review examines contemporary methodological approaches for investigating the human mitogenome. We discuss recent advancements in extraction and enrichment techniques, emphasizing their efficacy in reducing the interference of nuclear mitochondrial DNA sequences (NUMTs) and enhancing the recovery of informative fragments. Moreover, the shift from traditional Sanger sequencing to Massive Parallel Sequencing (MPS) is examined, as MPS has markedly enhanced the sensitivity and capability of contemporary methods to detect low-frequency heteroplasmies. Additionally, the advent of Third-Generation Sequencing (TGS), exemplified by nanopore platforms, is evaluated, which facilitates the reading of full-length native molecules without the biases introduced by PCR amplification. Despite the interpretive challenges posed by heteroplasmy, contamination, and limitations in population databases, ongoing methodological advances in mitochondrial DNA analysis continue to strengthen its reliability and expand its potential in forensic genetics. Full article

Estimating the post-mortem interval (PMI) remains a major challenge in forensic pathology, particularly beyond the earliest postmortem phases. RNA-based markers measured by RT-qPCR have been proposed as potential tools for PMI estimation, but their reliability depends on technical robustness, reference gene stability, and data representation. Technical reproducibility was overall satisfactory. However, the candidate reference genes showed different temporal behaviors: ACTB remained relatively stable, whereas RPL4 displayed significant time-dependent drift, resulting in partial instability of the composite reference signal. All target genes were associated with PMI, with HPRT1 and HMOX1 emerging as the most informative markers. Several targets also showed evidence of non-linear temporal dynamics. In predictive analyses, models based on raw Ct values consistently outperformed ΔCt-based models. A parsimonious model based on HMOX1 and HPRT1 showed the most favorable trade-off between interpretability and predictive performance within this exploratory dataset, although prediction error remained non-negligible. These findings suggest that postmortem cardiac transcriptional profiles may contain temporal information useful for PMI-oriented modeling but also show that predictive performance remains limited by reference gene behavior, analytical strategy, and non-negligible estimation error. Nine porcine hearts were stored at 4 °C and sampled at 0, 12, 24, 48, 72, 96, and 120 h. RT-qPCR was performed in technical triplicate for selected target genes (BAX, CASP3, HIF1A, HMOX1) together with additional quantified transcripts (ACTB, RPL4, GAPDH, HPRT1). Technical reproducibility, reference gene stability, temporal trends and predictive performance were assessed using mixed-effects models and predictive models evaluated by leave-one-heart-out cross-validation. Comparative analyses were performed using raw Ct, ΔCt, and ΔΔCt data. Overall, postmortem cardiac RT-qPCR profiling should be regarded as a proof-of-concept framework developed under specific controlled refrigerated conditions. Therefore, further external validation under heterogeneous real-world forensic scenarios and methodological standardization are required before a real-life forensic application. Full article

Background: Stalking is a well-established and studied crime in the forensic field. Nevertheless, research on the psychological aspects of stalking behaviors remains limited due to the lack of specific assessment tools. This two-phase research project aims to propose and validate a new instrument—the Stalking Assessment Questionnaire (SAQ)—to identify the possible latent psychological dimensions characterizing early stalking tendencies. Methods: During the first phase of questionnaire development, a thematic focus group comprising five psychologists generated 55 items. These items underwent preliminary screening with a sample of 85 students to assess clarity and redundancy; this process addressed semantic issues and resulted in a revised 48-item version. In the second validation phase (Study 1), the 48-item SAQ was administered to a sample of 349 male students whose demographic profiles matched those typically associated with potential stalkers; following this, 15 items were retained. In Study 2, the 15-item SAQ was cross-validated on an independent sample of 380 male students. To assess its validity, correlations were analyzed with a battery of personality inventories, including the LSRP, ECR-R, AQ, PESES, SRSS, RelRQ, ASI-S, and BIDR-6. Results: Following item and factor analyses, a model comprising two latent factors—Insistence (SAQ-INS) and Exaggerated Jealousy (SAQ-EXJ)—and 15 indicators emerged. Internal consistency was robust, with a Cronbach’s alpha of 0.87. In Study 2, as hypothesized, SAQ-EXJ correlated positively with anger and hostility (AQ) and with antisocial and impulsive traits (LSRP-F2), although these associations were modest (r = 0.30). Among the significant Pearson’s correlations, three were found to be substantial (r > 0.50): specifically, those between the SAQ and anxious attachment style (ECR-Ranx), relational rumination (RelRQ), and sexism (ASI-S). Conclusions: In conclusion, the final SAQ structure revealed two primary factors—Insistence and Exaggerated Jealousy—that account for stalking as a continuous construct. These factors demonstrated significant associations with several critical personality traits. The final questionnaire comprises 24 items, including nine fillers; consequently, it is efficient to administer and suitable for both research and psycho-educational interventions among adolescents and other potentially vulnerable populations. Full article