Objectives To develop and validate new regression equations to estimate the age of children and adolescents in a Yemeni population based on mandibular measurements. Materials & Methods This analytical cross-sectional study was conducted in two phases: the first phase included 338 individuals (141 females and 197 males), whose age ranged between 1 and 20 years. All mandible images were obtained using a 128-slice multidetector CT scanner. Seven mandibular dimensions were measured using the RadiAnt DICOM Viewer program. Linear regression was performed to develop age estimation equations. The second phase included a separate sample composed of 60 CT scans to validate and test the accuracy of the developed equations. Results The study demonstrated significant positive correlations between mandibular measurements and chronological age ranged from 0.788 for the chin height to 0.959 for the right ramus height (all p < 0.01). The single-dimension regression equations demonstrated high predictive power with determination coefficient ranging from 0.628 to 0.919. The multiple regression equations demonstrated the best overall performance and yielded the highest determination coefficients (R2=0.946 for males and 0.898 for females). Validation results indicated that the multiple regression equation yielded the lowest mean absolute error of 0.975 years (males) and 1.167 years (females), with the highest overall accuracy of 87.75% and 83.27%, respectively. Conclusions CT-based mandibular measurements are reliable indicators for estimating children and adolescents age. These findings offer a powerful tool for forensic, clinical, and anthropological applications within the Yemeni population. Further external validation is encouraged to enhance model applicability in diverse populations.
Decomposition changes the physical state and, thus, the imaging appearance of the body. Hallmarks of early decomposition are gas in the right heart and portal veins. When gas distribution is atypical, pathological conditions should be considered. Blood products can remain evident on PMCT even in moderate to advanced states of decomposition. As decomposition progresses, gas and fluid collections occur in all tissues and cavities, limiting diagnostic power. However, PMCT remains highly sensitive for the detection of skeletal injuries. Experienced interpreters are required to accurately differentiate between decomposition changes and pathology. The Guidelines Working Group of the International Society of Forensic Imaging endorses the application of non-contrasted PMCT for decomposed remains.
Introduction : Estimating sex from skeletal remains is a critical step in forensic anthropology. Traditional methods for visual assessment and measurement of sexually dimorphic characteristics rely on the skull, but they are susceptible to observer variability and are time-consuming. Recent advances in deep learning, particularly convolutional neural networks (CNNs), offer an opportunity to automate sex estimation directly from radiographic images, thereby enhancing efficiency and accuracy. Methods : This study compares the performance of three established CNN architectures; ResNet-18, ResNet-34, and DenseNet121 in classifying skull and neck radiographs as male or female. The models were trained and tested using cross-validation and automated hyperparameter optimization for robust model tuning. To enhance model interpretability, Gradient-weighted Class Activation Mapping (Grad-CAM) was applied to visualize the anatomical regions influencing classification decisions. Results All CNN models achieved accuracies of 85% or higher across both lateral and AP views, with DenseNet-121 (AP, weighted random sampling) achieving 95% in both accuracy and macro-F1 scores. ResNet-18 (lateral, 85% accuracy) and ResNet-34 (AP, 90% accuracy) trained with weighted random sampling demonstrated the lowest sex bias (2% and 6%, respectively), making them most suitable for fairness-critical forensic applications. Grad-CAM visualisations showed that all CNN models accurately identified major sexually dimorphic characteristics such as the glabella, supraorbital ridges, mastoid process, nuchal crest, mental eminence, cervical and mandible, demonstrating their ability to identify anatomically relevant regions. Conclusion : These findings demonstrate CNNs' potential to improve existing forensic approaches by providing scalable, objective, and automated solutions for sex estimation in forensic investigations and anthropological studies.
This study evaluates vascular opacification using extracorporeal membrane oxygenation (ECMO) cannulas for multiphase postmortem CT angiography (MPMCTA) in forensic cases.Five veno-arterial ECMO cases were selected, and vascular assessments were performed using seven arterial and five venous control points. ECMO cannulas were already in place upon the arrival of the body in the forensic center, inserted ante-mortem by medical staff during hospital care. Three parameters were evaluated: procedure time, procedural invasiveness and the quality of vascular opacification.The use of ECMO cannulas eliminated the need for standard femoral vessel exposure and cannulation, reducing setup time from 30 to 40 minutes to approximately 10 minutes. This approach also minimized invasiveness by avoiding the additional incisions typically required for femoral access. The vascular opacification quality achieved with ECMO cannulas was comparable to that of conventional MPMCTA.In conclusion, the use of pre-existing ECMO cannulas for MPMCTA is faster, less invasive and ensures reliable opacification quality comparable to the classical MPMCTA approach reported in the literature. This technique represents a practical and efficient alternative in cases where ECMO cannulas are already in place, improving workflow efficiency while preserving body integrity and diagnostic utility.
The use of advanced scanning technologies (ASTs) is advantageous for many aspects of the criminal investigation process, including documenting, assessing, and preserving evidence. This pilot project focuses on the application of reflective imaging modalities, specifically photogrammetry and structured light scanning (SLS), to pig mandibles (as proxies for humans), at various stages of decomposition. This research aims to determine if these technologies can be utilised in documenting remains at differing stages of decomposition in an outdoor crime scene scenario. This is achieved by assessing the accuracy of the resulting models created using the technologies at all stages of decomposition, and the feasibility of using these technologies in a simulated outdoor crime scene environment. The results highlight that both photogrammetry and SLS can create accurate models of remains in advanced decomposition and skeletonisation; however, issues were encountered when imaging remains in the early stages of decomposition. Photogrammetry was able to accurately recreate the colour and texture information of the fresh remains, where SLS could not. The project found that both photogrammetry and SLS are easy to use in the field; however, the SLS model used required additional equipment that created logistical complications. The results can assist in indicating which AST to employ to image remains in various stages of decomposition.
Traumatic brain injury is a cause for mortality and morbidity and a common occurrence in India where a Forensic medicine doctor deals with head injury cases every day. With rise of Forensic radiology as a specialty, the integration of computed tomography in autopsy has yielded positive results and especially when it is used as an adjuvant, the diagnostic accuracy of the entire procedure is better. In cases of head injury with skull fractures, tension pneumocephalus is a grave indicator which left untreated results in death of the individual. Radiologically, tension pneumocephalus is identified by a distinct “Mount Fuji” sign which holds true even for post-mortem computed tomography where the sign is clearly demonstrated. However, decomposition artefact also appears as a subdural gas collection in the frontal area which might lead to false reporting of a tension pneumocephalus. This paper aims to highlight this important pathology, the ability to differentiate it from an artefact and to motivate forensic medicine personnel to get trained in forensic radiology as use of such advanced modalities will overall improve the diagnostic accuracy of autopsies.
Background Venous lengths are a critical factor in selecting optimal central venous catheters. This study aimed to determine normative data for the superior vena cava (SVC) and its associated segments, including the SVC to axillary vein. Centerline vessel measurements, derived from realistic 3D geometries, involve tracing a curved line along the central axis of blood vessels using computer algorithms enabling precise evaluation of length and diameter. Methods Radiological imaging data were obtained from multiphase postmortem CT-angiography for the different SVC segments. Centerline measurements were analyzed to obtain normative values, compare them with anthropomorphic landmarks, and realize prediction models. Results Between January 2018 and October 2020, 219 subjects were retrospectively included (ethical committee registration BASEC-2021-00924). Mean centerline length ± standard deviation (in mm) was 64 ± 11 for the SVC; 75 ± 17, 30 ± 7, and 47 ± 10 for the axillary, subclavian, and brachiocephalic right veins, respectively; and 90 ± 15, 34 ± 7, and 76 ± 17 for the innominate, subclavian, and axillary left veins. The coefficient of variation was around 20%. The sums of the right and left sides were significantly different (216 ± 24 vs. 265 ± 26, p<0.001), only the right side corresponded with anthropomorphic landmarks (no significant difference). A prediction model based on height, sex, and age yielded two regression equations: Right [mm] = 57+0.8*height+11.1*sex+0.3*age; Left [mm] = 104+0.7*height+13.7*sex+0.6*age. Conclusion The results showed good to acceptable inter-subject variation in SVC segment lengths. Height, sex, and age were identified as moderate predictors of vessel length.
Traditional 2D mugshots have long been a cornerstone in criminal identification and forensic investigations. However, their limited dimensionality and reliance on cooperative subjects restrict their utility, especially when compared to oblique images captured by CCTV. Recent advancements in 3D imaging offer new possibilities for capturing detailed facial data, addressing many of the limitations inherent in 2D mugshot photography. This paper presents a fully developed 3D mugshot system based on photogrammetry, capable of generating high-resolution, texture-rich 3D models for police and forensic use, including standard forensic mugshots in high quality. Developed with practical requirements in mind, the system integrates seamlessly into existing forensic workflows, ensuring reliable documentation regardless of skin tone, height, or subject compliance. Key components include a 180-degree camera arc, synchronized with flash lighting. A height adjustable chair enables efficient capture with minimal operator training. The system has been deployed in routine criminal documentation, with about 7000 annual scans. This paper discusses the design, technical innovations, and benefits of this 3D mugshot system, as well as its potential for enhancing forensic analyses through applications in virtual and augmented reality. Additionally, the limitations of the system, including cost and data volume considerations, are examined. By bridging the gap between 2D and 3D imaging, this technology provides a significant advancement in the forensic documentation and identification of individuals.
Objectives EfficientNet is an up-to-date deep learning (DL) architecture and an important computer-based algorithm for forensic and basic medical sciences. The aim of this study is to perform highly accurate and precise sex prediction with images obtained from direct pelvic radiographs (X-Ray) using the EfficientNet deep learning model. Materials and Methods The study was performed using pelvis X-Ray images of 423 individuals aged 18-65 years. For the reliability of the DL architecture, data augmentation was applied to the images and the data set was increased to 1400. EfficientNet V2 B0 architecture was applied to the increased data set. The performance of the architecture was also compared using NASNet Mobile and ResNet152 rooted architectures. Results As a result of the study, a sex prediction rate of 99.29% was obtained with the EfficientNet architecture, 97.86% with the NASNet Mobile architecture and 95.71% with the ResNet152 architecture. Conclusion As a result of the study, EfficientNet V2 B0, NASNet Mobile and ResNet152 DL architectures showed high accuracy and reliability in terms of sex prediction. In this respect, we believe that this study will guide forensic and basic medical sciences.
Identifying deceased individuals using post-mortem computed tomography (PMCT) is a crucial challenge in forensic medicine. The aim of this study is to evaluate the feasibility of Fish&CheckNet, a novel Deep Learning (DL) model specifically designed for forensic identification through the automatic comparison of ante-mortem (AMCT) and PMCT facial scans. A retrospective feasibility study was conducted on 50 pairs of AMCT-PMCT scans acquired between 2022 and 2025. Fish&CheckNet uses a 3D Siamese architecture based on a modified ResNet-18 3D backbone. The model extracts biometric “fingerprints” (embeddings) from each CT volume and uses a “Matching Head” and a hybrid loss function to calculate the probability of a match. The model was trained on 40 pairs and validated on an independent test set of 20 pairs (10 positive and 10 negative). For comparison with expert practice, the same test set was blinded and analysed by an expert forensic imaging radiologist. On the test set, Fish&CheckNet achieved a ROC-AUC of 0.90 (CI 95%: 0.74-1.00). The overall accuracy of the model was 75% (CI 95%: 0.55-0.95), with a Sensitivity (Recall) of 0.80 and a Specificity of 0.70. The expert radiologist achieved an accuracy of 90% (18/20), demonstrating perfect specificity (1.00). Both recorded the same sensitivity (0.80). Fish&CheckNet demonstrates the potential of DL models to support forensic experts. Given the model's high false positive rate, it is currently best suited as a reliable pre-screening tool to quickly exclude low-probability non-matches, thereby simplifying the expert's workflow.
BACKGROUND : Identification of unknown individuals is a fundamental role of forensic anthropology, particularly in contexts where traditional methods such as DNA, fingerprints, and dental records are unavailable. Cephalometry is an alternative contemporary technique used to estimate sex, having high classification accuracy compared to conventional osteometric methods. This is the first study aimed to explore sexual dimorphism in a South African Black sample of cone beam computed tomography (CBCT)-derived lateral cephalograms. METHODS : This cross-sectional quantitative study analysed 79 CBCT-derived lateral cephalograms of adult Black South Africans. Lateral cephalograms were reconstructed in Avizo software. Ten 2D landmarks were identified and 12 linear distances derived. Univariate tests, as well as multivariate discriminant function analysis (direct and jackknifed) on linear distances and 2D landmarks were performed using Past software. RESULTS : Linear distances were larger in males than females for all variables and significant for Ba-Pr, Ba-N, S-N, N-ANS, ANS-PNS, S-Ba, A-point to B-point, A-point to Go, B-point to Go. Discriminant analysis of six variables with greater loadings achieved 72.15% jackknifed classification accuracy while four key 2D landmarks (S, Me, Go, Ba) yielded a 74.68% jackknifed classification accuracy. Wireframe analysis confirmed male mandibles as larger and more anteriorly projecting, with a more anteriorly positioned nasion. DISCUSSION AND CONCLUSION : CBCT-derived lateral cephalograms effectively demonstrate sexual dimorphism in a Black South African population. Classification accuracies improved when focusing on variables with the highest discriminative power. These findings highlight the potential of cephalometric techniques as complementary tools for forensic sex estimation, with implications for building population-specific reference databases.
Introduction: Determining the degree of visible decomposition is crucial in the forensic investigation of decomposed cases. Forensic pathologists traditionally conduct gross assessment of the decomposition stage, alongside other influencing factors, to narrow the estimated postmortem interval (PMI). With advances in forensic radiology, postmortem computed tomography (PMCT) has become an important adjunctive tool providing additional internal information. During decomposition, the brain undergoes autolysis and settles in gravity-dependent areas, and putrefaction gas accumulation, which is detectable by PMCT, occurs in nondependent areas. Objective: To analyze the relationship between the external appearance grade (EAG) of decomposed cases in Thailand and the persisting decomposed intracranial content percentage (PICP) calculated from PMCT data. Research method: This retrospective study included 175 decomposed cases, focusing on EAG 0-3, examined at the Forensic Department, Faculty of Medicine, Chulalongkorn University, for which PMCT was performed between May 2022 and February 2025. Cases were categorized into four grades based on photographs taken around the same time as PMCT. The data were then processed using 3D Slicer software to calculate PICP from voxel-based analysis. Results and conclusion: Spearman's correlation analysis revealed a moderate negative correlation between PICP and EAG across grades 0-3 (correlation coefficient = -0.728, p < 0.001). Ordinal logistic regression found that sex and age were nonsignificant parameters, whereas PICP remained significantly associated (p < 0.001) with EAG 0-3. This study highlighted the utilization of radioinformatics for investigating the relationship between objective internal findings and external decomposition appearance in decomposed bodies.
Purpose: The aim of this study was to evaluate the utility of different heart wall measurement positions in PMCT and PMMR and to determine the accuracy of heart wall measurements in PMCT compared to PMMR. Materials and methods: We retrospectively reviewed all cases of individuals over 18 years old, who had received a PMCT and short axis cardiac PMMR. Exclusion criteria included extensive gas accumulation, poor PMMR image quality due to protocol limitations and undetectable left ventricular lumen in PMCT or PMMR. In total, a maximum of 9 measurements were taken per case and per modality, and the measurements were repeated after 10 days to evaluate intrarater agreement. Results: A total of 102 cases were included in the study. Generally, the agreement was higher for measurements of the anterior wall compared to the free wall and the septum, and higher for measurements at the apex compared to the base of the heart. Additionally, the agreement within a modality was higher in PMMR than PMCT. Comparing the measurements of the two methods, we found the anterior wall thickness in PMCT to be the most precise compared to PMMR. Conclusion: Left ventricular wall measurements show a higher agreement in PMMR compared to PMCT. Measurements of the anterior wall of the left ventricle showed the least difference between PMCT and PMMR and had the highest intrarater agreement.
A 53-year-old man was found dead in a supine position on a sofa. A postmortem computed tomography (PMCT) and forensic autopsy were performed 4 days after his death. No external injuries were observed. The heart weighed 463 g and showed mild hypertrophy without gross myocardial infarction. A 1.3-cm intimal tear was identified approximately 7 cm above the aortic valve, between the left common carotid and subclavian arteries. A Stanford type A aortic dissection extending from the aortic root to the bifurcation of the common iliac arteries, involving the left coronary artery, without adventitial rupture or intrathoracic hemorrhage was observed. Fusion imaging of histopathological findings with three-dimensional (3D)-reconstructed PMCT images clearly visualized the dissection extending into the coronary artery. The patient’s cause of death was acute aortic dissection with secondary myocardial ischemia due to impaired coronary blood flow. To the best of our knowledge, this is the first report is the first to demonstrate coronary involvement by combining histology with 3D PMCT reconstruction.
Sex estimation is the first biological characteristic investigated while identifying an individual. Methods using different technologies, such as three-dimensional (3D) images, provide a new means of applying forensic methodologies. The mandible is known for its sexual dimorphic potential, being widely used in research related to forensic anthropology. This research aims to describe, analyze, and synthesize the most relevant research on human sex estimation based on anthroposcopic and anthropometric analyses in 3D images generated from computed tomography (CT) scans of dry mandibles of postmortem individuals. This is a scoping review, with a description and analysis of the literature. The development of the review followed the standards defined and recommended by the Joanna Briggs Institute Manual (JBI Manual for Evidence Synthesis) and the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses for Scoping Reviews (PRISMA-ScR). The protocol was registered on the Open Science Framework under registration: DOI 10.17605/OSF.IO/CW8MV. The study used MEDLINE databases via PubMed, Web of Science, and Scopus, as well as gray literature via Google Scholar. Altogether, 564 articles were obtained, and 33 were selected to be read in full after the screening process. Lastly, two articles were included in the scoping review. Both articles concluded that the mandible can be used for sex estimation, using CT scan images. The results of the studies varied because they had different focuses. One sought to validate the use of mandibular images for sex estimation, obtaining an 85.5% accuracy rate. The other study sought to validate the applicability of an instrument to measure the mental protuberance correctly, concluding that even forensic anthropologists with extensive experience need to undergo in-depth radiological studies to use digital images to estimate sex. This study concluded that access to collections of dry mandibles is limited due to each country’s national legislation and that digitizing collections would enable more forensic anthropology studies.
This study aimed to evaluate the feasibility and accuracy of an artificial intelligence (AI)-based method for automatically measuring femoral length from postmortem computed tomography (PMCT) images and to develop regression models for estimating stature in Japanese individuals. PMCT data from 163 deceased individuals aged 20–73 years were examined using an AI-based segmentation tool. Semantic segmentation was performed using the TotalSegmentator library, and the maximum femoral length was estimated using the Double Sweep method. Measurement reproducibility was assessed using intraclass correlation coefficients (ICC). The adjusted stature (AS) was calculated by subtracting 2.0 cm from the stature measured during the autopsy to estimate living stature. The correlations between femoral measurements and AS were analyzed using simple linear regression. After deriving the stature estimation formulae from the regression analyses, the time required from the initiation of femoral measurement to the display of the estimated stature was recorded. Femoral measurements demonstrated excellent reproducibility, with ICC values > 0.999 and no outliers detected. Significant positive correlations were observed between femoral measurements and AS across all models (p < 0.001). The adjusted coefficients of determination values exceeded 0.80 for the total sample, and the root mean square error values were 3.5 cm or less. The mean time from the initiation of femoral measurement to the display of the estimated stature was 23.05 s. AI-driven femoral measurement from PMCT images provides a highly accurate, efficient, and reproducible approach for forensic stature estimation. Not applicable.
This study presents PanDentNet, a lightweight shared-backbone multi-task convolutional network with three task-specific heads for dental radiograph analysis, covering caries burden stratification, periapical anomaly classification, and restoration classification. In contrast to conventional single-task pipelines, PanDentNet is trained across heterogeneous datasets in a partially labeled multi-task setting, where label masking ensures that only available annotations contribute to the corresponding task losses while a common representation is learned in the shared backbone. Without using data augmentation, PanDentNet achieves accuracies of 73.49%, 68.50%, and 85.12% on the Approximal Caries Periapical Dataset (ACPD), Tufts, and UFBA-UESC datasets, respectively. Under an identical ACPD validation split, an ablation study shows that the multi-task formulation improves the primary caries task over a single-task CNN baseline (73.49% vs. 67.47% accuracy; 0.81 vs. 0.78 F1), suggesting positive transfer from related radiographic tasks. In summary, the findings indicate that a compact multi-head model can support unified decision-making for dental radiograph screening and may be relevant to both clinical and forensic applications.
Photon-counting detector (PCD) computed tomography (CT) offers reduced image noise, high spatial resolution, and intrinsic spectral imaging capabilities, making it a promising technology for post-mortem CT. However, the relative contribution of PCD-CT to post-mortem imaging still needs to be characterized. This short communication provides a stepwise evaluation of key determinants of image quality in PMCT using a controlled post-mortem animal model. Whole-body CT was performed on a single post-mortem animal specimen using a PCD CT system and a conventional energy-integrating detector (EID) CT system under matched acquisition and reconstruction conditions. Image quality was evaluated sequentially by comparing background noise characteristics between detector types, assessing high-resolution photon-counting CT protocols for osseous structures, and analyzing soft-tissue image quality using SNR and CNR under both polychromatic and virtual mono-energetic reconstruction conditions. PCD CT demonstrated significantly lower background noise compared with EID CT (p < 0.01). High-resolution PCD CT markedly improved visualization of fine osseous and pulmonary structures, while no consistent qualitative benefit for soft tissue was visually observed. Under matched polychromatic conditions, soft-tissue contrast improvements were modest and tissue-dependent. The most pronounced soft-tissue advantages were observed with spectral imaging, where virtual mono-energetic reconstructions showed a strong energy-dependent behavior. CNR improvements were organ-specific, with clear benefits for renal parenchyma but not for liver tissue. These findings indicate that the principal added value of PCD CT in post-mortem imaging is task-dependent. Detector-related noise reduction provides a robust baseline benefit, high-resolution imaging primarily enhances osseous assessment, and spectral reconstructions offer the greatest potential for soft-tissue optimization.
