Dental age estimation is a critical part of forensic odontology applications. It is conventionally carried out by experts on dental orthopantomograms manually. Deep learning models are the current state-of-the-art in automated dental age estimation applications. When applying deep learning models to high-stakes scenarios, such as forensic investigations, understanding the model’s behaviour is critical. In this context, we investigate a scenario where the automated staging of tooth 37 and tooth 38, the left second and third mandibular molars, with a vision transformer (ViT) model displays a striking performance disparity, achieving accuracies of 0.64 and 0.37, respectively. We further explore the self-attention mechanism of the ViT in search of the reason for this disparity. Furthermore, to introduce additional transparency, we propose a pipeline consisting of a convolutional autoencoder (AE) trained with triplet loss, and a ViT model for classification. This pipeline achieves accuracies of 0.72 and 0.39 on teeth 37 and 38. We further explore the latent space, image reconstructions and attention maps, contrasting the two teeth in order to uncover the reason for the poor performance on tooth 38. Through probing our pipeline, we reveal the problem of high intra-class variation in the tooth 38 dataset, and demonstrate an approach that can offer increased transparency in deep learning applications on medical image data.

The practical adoption of deep learning in high-stakes forensic applications, such as dental age estimation, is often limited by the 'black box' nature of the models. This study introduces a framework designed to enhance both performance and transparency in this context. We use a notable performance disparity in the automated staging of mandibular second (tooth 37) and third (tooth 38) molars as a case study. The proposed framework, which combines a convolutional autoencoder (AE) with a Vision Transformer (ViT), improves classification accuracy for both teeth over a baseline ViT, increasing from 0.712 to 0.815 for tooth 37 and from 0.462 to 0.543 for tooth 38. Beyond improving performance, the framework provides multi-faceted diagnostic insights. Analysis of the AE's latent space metrics and image reconstructions indicates that the remaining performance gap is data-centric, suggesting high intra-class morphological variability in the tooth 38 dataset is a primary limiting factor. This work highlights the insufficiency of relying on a single mode of interpretability, such as attention maps, which can appear anatomically plausible yet fail to identify underlying data issues. By offering a methodology that both enhances accuracy and provides evidence for why a model may be uncertain, this framework serves as a more robust tool to support expert decision-making in forensic age estimation.

Accurate functional outcome prediction shortly after the onset of stroke would enable more effective personalized care of stroke patients. A deep learning approach was developed that utilizes follow-up images at 24 h to predict the functional outcome 90 days after stroke onset. The method involves the use of a conventional U-net segmentation model that was trained to delineate the stroke lesion on CT images, with an additional branch integrated into the U-net’s bottom layer to extract features for a separately trained network that classifies cases into favorable (modified Rankin Score (mRS) = 0–2) or unfavorable outcomes (mRS = 3–6). The method was trained and validated using 3-fold cross-validation on a set of 240 images (training: 170; validation: 90). The lesion segmentation yielded an average Dice score of 0.458 and an average absolute volume error of 10.71 ml, while the binary mRS prediction obtained an overall accuracy of 88.6

Three-dimensional (3D) anatomical extraction techniques could help the forensic anthropologist in a precise and inclusive assessment of biological phenotypes for the development of facial reconstruction methods. In this research, the nose morphology and the underlying hard tissue of two South African populations were studied. To this end, a 3D computer-assisted approach based on an automated landmarking workflow was used to generate relevant 3D anatomical components, and shape discrepancies were investigated using a data set of 200 cone-beam computer tomography (CBCT) scans. The anatomical landmarks were placed on the external nose and the mid-facial skeleton (the nasal bones, the anterior nasal aperture, the zygoma, and the maxilla). Shape differences related to population affinity, sex, age, and size were statistically evaluated and visualised using geometric morphometric methods. Population affinity, sexual dimorphism, age, and size affect the nasal complex morphology. Shape variation in the mid-facial region was significantly influenced by population affinity, emphasising that shape variability was specific to the two population groups, along with the expression of sexual dimorphism and the effect of ageing. In addition, nasal complex shape and correlations vary greatly between white and black South Africans, highlighting a need for reliable population-specific 3D statistical nose prediction algorithms.

Age estimation in forensic odontology is mainly based on the development of permanent teeth. To register the developmental status of an examined tooth, staging techniques were developed. However, due to inappropriate calibration, uncertainties during stage allocation, and lack of experience, non-uniformity in stage allocation exists between expert observers. As a consequence, related age estimation results are inconsistent. An automated staging technique applicable to all tooth types can overcome this drawback. This study aimed to establish an integrated automated technique to stage the development of all mandibular tooth types and to compare their staging performances. Calibrated observers staged FDI teeth 31, 33, 34, 37 and 38 according to a ten-stage modified Demirjian staging technique. According to a standardised bounding box around each examined tooth, the retrospectively collected panoramic radiographs were cropped using Photoshop CC 2021® software (Adobe®, version 23.0). A gold standard set of 1639 radiographs were selected (n31 = 259, n33 = 282, n34 = 308, n37 = 390, n38 = 400) and input into a convolutional neural network (CNN) trained for optimal staging accuracy. The performance evaluation of the network was conducted in a five-fold cross-validation scheme. In each fold, the entire dataset was split into a training and a test set in a non-overlapping fashion between the folds (i.e., 80

PurposeFacial disfigurement may affect the quality of life of many patients. Facial prostheses are often used as an adjuvant to surgical intervention and may sometimes be the only viable treatment option. Traditional methods for designing soft-tissue facial prostheses are time-consuming and subjective, while existing digital techniques are based on mirroring of contralateral features of the patient, or the use of existing feature templates/models that may not be readily available. We aim to support the objective and semi-automated design of facial prostheses with primary application to midline or bilateral defect restoration where no contralateral features are present. Specifically, we developed and validated a statistical shape model (SSM) for estimating the shape of missing facial soft tissue segments, from any intact parts of the face.Materials and MethodsAn SSM of 3D facial variations was built from meshes extracted from computed tomography and cone beam computed tomography images of a black South African sample (n = 235) without facial disfigurement. Various types of facial defects were simulated, and the missing parts were estimated automatically by a weighted fit of each mesh to the SSM. The estimated regions were compared to the original regions using color maps and root-mean-square (RMS) distances.ResultsRoot mean square errors (RMSE) for defect estimations of one orbit, partial nose, cheek, and lip were all below 1.71 mm. Errors for the full nose, bi-orbital defects, as well as small and large composite defects were between 2.10 and 2.58 mm. Statistically significant associations of age and type of defect with RMSE were observed, but not with sex or imaging modality.ConclusionThis method can support the objective and semi-automated design of facial prostheses, specifically for defects in the midline, crossing the midline or bilateral defects, by facilitating time-consuming and skill-dependent aspects of prosthesis design.

Bloodstain pattern analysis plays a crucial role in forensic investigations. Projected patterns can offer valuable insights into the dynamics of crime scenes. In this paper, we propose and validate a novel approach that extends existing software, HemoVision, to analyze impact patterns that are distributed across multiple arbitrarily oriented surfaces. The proposed method integrates HemoVision’s marker-based system with structure from motion (SfM) techniques to reconstruct the three-dimensional geometry of impact patterns using only two-dimensional photographs. Controlled experiments were used to validate the proposed approach, demonstrating robustness in reconstruction accuracy with median translation errors below 3 mm and median angular errors below 0.2°, irrespective of imaging device or image resolution. Comparing the estimated areas origin to their known ground truth, the proposed method achieved an average total error of 8.12 cm, with the primary source of error being the vertical dimension. Despite this, the overall error remains well within the ranges of error reported in prior work. This study demonstrates that HemoVision can be used to analyze complex impact patterns using only two-dimensional photographs, providing forensic experts with an efficient and accessible tool for investigating intricate crime scenes involving multi-surface impact patterns.

Dental age estimation, a cornerstone in forensic age assessment, has been extensively tried and tested, yet manual methods are impeded by tedium and interobserver variability. Automated approaches using deep transfer learning encounter challenges like data scarcity, suboptimal training, and fine-tuning complexities, necessitating robust training methods. This study explores the impact of convolutional neural network hyperparameters, model complexity, training batch size, and sample quantity on age estimation. EfficientNet-B4, DenseNet-201, and MobileNet V3 models underwent cross-validation on a dataset of 3896 orthopantomograms (OPGs) with batch sizes escalating from 10 to 160 in a doubling progression, as well as random subsets of this training dataset. Results demonstrate the EfficientNet-B4 model, trained on the complete dataset with a batch size of 160, as the top performer with a mean absolute error of 0.562 years on the test set, notably surpassing the MAE of 1.01 at a batch size of 10. Increasing batch size consistently improved performance for EfficientNet-B4 and DenseNet-201, whereas MobileNet V3 performance peaked at batch size 40. Similar trends emerged in training with reduced sample sizes, though they were outperformed by the complete models. This underscores the critical role of hyperparameter optimization in adopting deep learning for age estimation from complete OPGs. The findings not only highlight the nuanced interplay of hyperparameters and performance but also underscore the potential for accurate age estimation models through optimization. This study contributes to advancing the application of deep learning in forensic age estimation, emphasizing the significance of tailored training methodologies for optimal outcomes.

ABSTRACT Vertebral fractures (VFs) are the hallmark of osteoporosis, being one of the most frequent types of fragility fracture and an early sign of the disease. They are associated with significant morbidity and mortality. VFs are incidentally found in one out of five imaging studies, however, more than half of the VFs are not identified nor reported in patient computed tomography (CT) scans. Our study aimed to develop a machine learning algorithm to identify VFs in abdominal/chest CT scans and evaluate its performance. We acquired two independent data sets of routine abdominal/chest CT scans of patients aged 50 years or older: a training set of 1011 scans from a non-interventional, prospective proof-of-concept study at the Universitair Ziekenhuis (UZ) Brussel and a validation set of 2000 subjects from an observational cohort study at the Hospital of Holbæk. Both data sets were externally reevaluated to identify reference standard VF readings using the Genant semiquantitative (SQ) grading. Four independent models have been trained in a cross-validation experiment using the training set and an ensemble of four models has been applied to the external validation set. The validation set contained 15.3% scans with one or more VF (SQ2-3), whereas 663 of 24,930 evaluable vertebrae (2.7%) were fractured (SQ2-3) as per reference standard readings. Comparison of the ensemble model with the reference standard readings in identifying subjects with one or more moderate or severe VF resulted in an area under the receiver operating characteristic curve (AUROC) of 0.88 (95% confidence interval [CI], 0.85–0.90), accuracy of 0.92 (95% CI, 0.91–0.93), kappa of 0.72 (95% CI, 0.67–0.76), sensitivity of 0.81 (95% CI, 0.76–0.85), and specificity of 0.95 (95% CI, 0.93–0.96). We demonstrated that a machine learning algorithm trained for VF detection achieved strong performance on an external validation set. It has the potential to support healthcare professionals with the early identification of VFs and prevention of future fragility fractures. © 2023 UCB S.A. and The Authors. Journal of Bone and Mineral Research published by Wiley Periodicals LLC on behalf of American Society for Bone and Mineral Research (ASBMR). Abstract Vertebral fractures are the landmark of osteoporosis and are associated with increased risk of further fractures. Yet they commonly go undiagnosed, and the underlying osteoporosis untreated. We developed a machine learning algorithm for automated vertebral fracture detection and demonstrated strong performance on an external validation set of 2000 CT scans. This method can improve the identification and reporting of vertebral fractures by opportunistically screening for them in routine CT scans.

The cranial vault in humans is highly variable, clinically relevant, and heritable, yet its genetic architecture remains poorly understood. Here, we conduct a joint multi-ancestry and admixed multivariate genome-wide association study on 3D cranial vault shape extracted from magnetic resonance images of 6772 children from the ABCD study cohort yielding 30 genome-wide significant loci. Follow-up analyses indicate that these loci overlap with genomic risk loci for sagittal craniosynostosis, show elevated activity cranial neural crest cells, are enriched for processes related to skeletal development, and are shared with the face and brain. We present supporting evidence of regional localization for several of the identified genes based on expression patterns in the cranial vault bones of E15.5 mice. Overall, our study provides a comprehensive overview of the genetics underlying normal-range cranial vault shape and its relevance for understanding modern human craniofacial diversity and the etiology of congenital malformations.

We know that both the CNN mapping function and the sampling scheme are of paramount importance for CNN-based image analysis. It is clear that both functions operate in the same space, with an image axis $\mathcal{I}$ and a feature axis $\mathcal{F}$. Remarkably, we found that no frameworks existed that unified the two and kept track of the spatial origin of the data automatically. Based on our own practical experience, we found the latter to often result in complex coding and pipelines that are difficult to exchange. This article introduces our framework for 1, 2 or 3D image classification or segmentation: DeepVoxNet2 (DVN2). This article serves as an interactive tutorial, and a pre-compiled version, including the outputs of the code blocks, can be found online in the public DVN2 repository. This tutorial uses data from the multimodal Brain Tumor Image Segmentation Benchmark (BRATS) of 2018 to show an example of a 3D segmentation pipeline.

Albeit the Dice loss is one of the dominant loss functions in medical image segmentation, most research omits a closer look at its derivative, i.e. the real motor of the optimization when using gradient descent. In this paper, we highlight the peculiar action of the Dice loss in the presence of missing or empty labels. First, we formulate a theoretical basis that gives a general description of the Dice loss and its derivative. It turns out that the choice of the reduction dimensions and the smoothing term ϵ is non-trivial and greatly influences its behavior. We find and propose heuristic combinations of and ϵ that work in a segmentation setting with either missing or empty labels. Second, we empirically validate these findings in a binary and multiclass segmentation setting using two publicly available datasets. We confirm that the choice of and ϵ is indeed pivotal. With chosen such that the reductions happen over a single batch (and class) element and with a negligible ϵ , the Dice loss deals with missing labels naturally and performs similarly compared to recent adaptations specific for missing labels. With chosen such that the reductions happen over multiple batch elements or with a heuristic value for ϵ , the Dice loss handles empty labels correctly. We believe that this work highlights some essential perspectives and hope that it encourages researchers to better describe their exact implementation of the Dice loss in future work.

In this article, we look into some essential aspects of convolutional neural networks (CNNs) with the focus on medical image segmentation. First, we discuss the CNN architecture, thereby highlighting the spatial origin of the data, voxel-wise classification and the receptive field. Second, we discuss the sampling of input-output pairs, thereby highlighting the interaction between voxel-wise classification, patch size and the receptive field. Finally, we give a historical overview of crucial changes to CNN architectures for classification and segmentation, giving insights in the relation between three pivotal CNN architectures: FCN, U-Net and DeepMedic.

Bloodstain impact pattern Area of Origin (AO) estimation is an important but time-consuming process in criminal investigations. HemoVision is a software package that automates and accelerates this process. To date, however, no study has been published that evaluates HemoVision's accuracy. Moreover, HemoVision relies on an automated variant of the tangent method to estimate a pattern's AO, even though the use of front-view projections has been shown to provide biased AO estimates. Therefore, the goal of this paper is twofold. First, a novel AO estimation method is proposed, whereby AO estimation is formulated as a least-squares optimisation problem that operates in three dimensions directly, eliminating the need for front view projections. Second, ten impact patterns with known AO coordinates at both 50 cm and 100 cm with respect to the target wall are created and used to compare the proposed approach's accuracy and robustness to the manual tangent method, HemoSpat, and HemoVision's automated tangent method. Results show that for impacts at 100 cm or less to the target wall, the proposed approach achieves the lowest average error of 17.29 cm with the least uncertainty, and that it performs significantly better than the manual tangent and automated tangent methods at a 5% significance level. Moreover, it is shown to achieve errors of less than 30 cm at these distances with just nine stains, whereas the automated tangent method requires a minimum of 16 stains to achieve the same average error. (C) 2022 Elsevier B.V. All rights reserved.

This article focuses on the control center of each human body: the brain. We will point out the pivotal role of the cerebral vasculature and how its complex mechanisms may vary between subjects. We then emphasize a specific acute pathological state, i.e., acute ischemic stroke, and show how medical imaging and its analysis can be used to define the treatment. We show how the core-penumbra concept is used in practice using mismatch criteria and how machine learning can be used to make predictions of the final infarct, either via deconvolution or convolutional neural networks.

Convolutional neural networks (CNNs) for brain tumor segmentation are generally developed using complete sets of magnetic resonance imaging (MRI) sequences for both training and inference. As such, these algorithms are not trained for realistic, clinical scenarios where parts of the MRI sequences which were used for training, are missing during inference. To increase clinical applicability, we proposed a cross-modal distillation approach to leverage the availability of multi-sequence MRI data for training and generate an enriched CNN model which uses only single-sequence MRI data for inference but outperforms a single-sequence CNN model. We assessed the performance of the proposed method for whole tumor and tumor core segmentation with multi-sequence MRI data available for training but only $T_{1}$ -weighted ( $T_{\text{1}w}$ ) sequence data available for inference, using BraTS 2018, and in-house datasets. Results showed that cross-modal distillation significantly improved the Dice score for both whole tumor and tumor core segmentation when only $T_{\text{1}w}$ sequence data were available for inference. For the evaluation using the in-house dataset, cross-modal distillation achieved an average Dice score of 79.04% and 69.39% for whole tumor and tumor core segmentation, respectively, while a single-sequence U-Net model using $T_{\text{1}w}$ sequence data for both training and inference achieved an average Dice score of 73.60% and 62.62%, respectively. These findings confirmed cross-modal distillation as an effective method to increase the potential of single-sequence CNN models such that segmentation performance is less compromised by missing MRI sequences or having only one MRI sequence available for segmentation.

A novel algorithm for generating artificial training samples from triangulated three-dimensional (3D) surface models within the context of dental implant recognition is proposed. The proposed algorithm is based on the calculation of two-dimensional (2D) projections (from a number of different angles) of 3D volumetric representations of computer-aided design (CAD) surface models. A fully convolutional network (FCN) is subsequently trained on the artificially generated X-ray images for the purpose of automatically identifying the connection type associated with a specific dental implant in an actual X-ray image. Semi-automated and fully automated systems are proposed for segmenting questioned dental implants from the background in actual X-ray images. Within the context of the semi-automated system, suitable regions of interest (ROIs), which contain the dental implants, are manually specified. However, as part of the fully automated system, suitable ROIs are automatically detected. It is demonstrated that a segmentation/detection accuracy of 94.0% and a classification/recognition accuracy of 71.7% are attainable within the context of the proposed fully automated system. Since the proposed systems utilise an ensemble of techniques that has not been employed for the purpose of dental implant classification/recognition on any previous occasion, the above-mentioned results are very encouraging.

The cranial vault – the portion of the skull surrounding the brain and cerebellum – is highly variable, clinically relevant, and heritable, yet its genetic architecture remains poorly understood. Here, we conducted a joint multi-ancestry and admixed multivariate GWAS on 3D cranial vault shape extracted from magnetic resonance images of 6,772 children from the ABCD study cohort, identifying 30 genome-wide significant genetic loci and replicating 20 of these signals in 16,947 additional individuals of the UK Biobank. This joint multi-ancestry GWAS was enriched for genetic components of cranial vault shape shared across ancestral groups and yielded a greater discovery than a European-only GWAS. We present supporting evidence for parietal versus frontal bone localization for several of the identified genes based on expression patterns in E15.5 mice. Collectively, our GWAS loci were enriched for processes related to skeletal development and showed elevated activity in cranial neural crest cells, suggesting a role during early craniofacial development. Among the identified genes, were RUNX2 and several of its upstream and downstream actors, highlighting the prominent role of intramembranous ossification – which takes place at the cranial sutures – in influencing cranial vault shape. We found that mutations in many genes associated with craniosynostosis exert their pathogenicity by modulating the same pathways involved in normal cranial vault development. This was further demonstrated in a non-syndromic sagittal craniosynostosis case-parent trio dataset of 63 probands (n = 189), where our GWAS signals near BMP2, BBS9 , and ZIC2 contributed significantly to disease risk. Moreover, we found strong evidence of overlap with genes influencing the morphology of the face and the brain, suggesting a common genetic architecture connecting these developmentally adjacent structures. Overall, our study provides a comprehensive overview of the genetics underlying normal cranial vault shape and its relevance for understanding modern human craniofacial diversity and the etiology of congenital malformations.