Classification of rib fracture types from postmortem computed tomography images using deep learning.

Human or time resources can sometimes fall short in medical image diagnostics, and analyzing images in full detail can be a challenging task. With recent advances in artificial intelligence, an increasing number of systems have been developed to assist clinicians in their work. In this study, the objective was to train a model that can distinguish between various fracture types on different levels of hierarchical taxonomy and detect them on 2D-image representations of volumetric postmortem computed tomography (PMCT) data. We used a deep learning model based on the ResNet50 architecture that was pretrained on ImageNet data, and we used transfer learning to fine-tune it to our specific task. We trained our model to distinguish between "displaced," "nondisplaced," "ad latus," "ad longitudinem cum contractione," and "ad longitudinem cum distractione" fractures. Radiographs with no fractures were correctly predicted in 95-99% of cases. Nondisplaced fractures were correctly predicted in 80-86% of cases. Displaced fractures of the "ad latus" type were correctly predicted in 17-18% of cases. The other two displaced types of fractures, "ad longitudinem cum contractione" and "ad longitudinem cum distractione," were correctly predicted in 70-75% and 64-75% of cases, respectively. The model achieved the best performance when the level of hierarchical taxonomy was high, while it had more difficulties when the level of hierarchical taxonomy was lower. Overall, deep learning techniques constitute a reliable solution for forensic pathologists and medical practitioners seeking to reduce workload.

Augmented reality in forensics and forensic medicine - Current status and future prospects.

Forensic investigations require a vast variety of knowledge and expertise of each specialist involved. With the increase in digitization and advanced technical possibilities, the traditional use of a computer with a screen for visualization and a mouse and keyboard for interactions has limitations, especially when visualizing the content in relation to the real world. Augmented reality (AR) can be used in such instances to support investigators in various tasks at the scene as well as later in the investigation process. In this article, we present current applications of AR in forensics and forensic medicine, the technological basics of AR, and the advantages that AR brings for forensic investigations. Furthermore, we will have a brief look at other fields of application and at future developments of AR in forensics.

Automated wound segmentation and classification of seven common injuries in forensic medicine.

In forensic medical investigations, physical injuries are documented with photographs accompanied by written reports. Automatic segmentation and classification of wounds on these photographs could provide forensic pathologists with a tool to improve the assessment of injuries and accelerate the reporting process. In this pilot study, we trained and compared several preexisting deep learning architectures for image segmentation and wound classification on forensically relevant photographs in our database. The best scores were a mean pixel accuracy of 69.4% and a mean intersection over union (IoU) of 48.6% when evaluating the trained models on our test set. The models had difficulty distinguishing the background from wounded areas. As an example, image pixels showing subcutaneous hematomas or skin abrasions were assigned to the background class in 31% of cases. Stab wounds, on the other hand, were reliably classified with a pixel accuracy of 93%. These results can be partially attributed to undefined wound boundaries for some types of injuries, such as subcutaneous hematoma. However, despite the large class imbalance, we demonstrate that the best trained models could reliably distinguish among seven of the most common wounds encountered in forensic medical investigations.

Metabolomics-based Sleepiness Markers for Risk Prevention and Traffic Safety (ME-SMART): a monocentric, controlled, randomized, crossover trial.

BACKGROUND: Too little sleep and the consequences thereof are a heavy burden in modern societies. In contrast to alcohol or illicit drug use, there are no quick roadside or workplace tests for objective biomarkers for sleepiness. We hypothesize that changes in physiological functions (such as sleep-wake regulation) are reflected in changes of endogenous metabolism and should therefore be detectable as a change in metabolic profiles. This study will allow for creating a reliable and objective panel of candidate biomarkers being indicative for sleepiness and its behavioral outcomes. METHODS: This is a monocentric, controlled, randomized, crossover, clinical study to detect potential biomarkers. Each of the anticipated 24 participants will be allocated in randomized order to each of the three study arms (control, sleep restriction, and sleep deprivation). These only differ in the amount of hours slept per night. In the control condition, participants will adhere to a 16/8 h wake/sleep regime. In both sleep restriction and sleep deprivation conditions, participants will accumulate a total sleep deficit of 8 h, achieved by different wake/sleep regimes that simulate real-life scenarios. The primary outcome is changes in the metabolic profile (i.e., metabolome) in oral fluid. Secondary outcome measures will include driving performance, psychomotor vigilance test, d2 Test of Attention, visual attention test, subjective (situational) sleepiness, electroencephalographic changes, behavioral markers of sleepiness, changes in metabolite concentrations in exhaled breath and finger sweat, and correlation of metabolic changes among biological matrices. DISCUSSION: This is the first trial of its kind that investigates complete metabolic profiles combined with performance monitoring in humans over a multi-day period involving different sleep-wake schedules. Hereby, we aim to establish a candidate biomarker panel being indicative for sleepiness and its behavioral outcomes. To date, there are no robust and easily accessible biomarkers for the detection of sleepiness, even though the vast damage on society is well known. Thus, our findings will be of high value for many related disciplines. TRIAL REGISTRATION: ClinicalTrials.gov Identifier NCT05585515, released on 18.10.2022; Swiss National Clinical Trial Portal SNCTP000005089, registered on 12 August 2022.

MRI Segmentation of Cervical Muscle Volumes in Survived Strangulation: Is There an Association between Side Differences in Muscle Volume and the Handedness of the Perpetrator? A Retrospective Study.

We evaluate the potential value of magnetic resonance imaging (MRI) in the examination of survivors of manual strangulation. Our hypothesis was that trauma-induced edema of the cervical muscles might lead to a side difference in the muscle volumes, associated with the handedness of the perpetrator. In 50 individuals who survived strangulation, we performed MRI-based segmentation of the cervical muscle volumes. As a control group, the neck MRIs of 10 clinical patients without prior trauma were used. The ratio of the right to left muscle volume was calculated for each muscle group of the control and strangulation groups. Cutoff values for the assumed physiological muscle volume ratios between the right and left sides were identified from our control group. There was no significant difference among the individuals in the pathological muscle volume ratio between right-handed versus both-handed strangulation for the sternocleidomastoid, pretracheal, anterior deep, or trapezoid muscle groups. Only the posterior deep muscle group showed a statistically significant difference in the pathological muscle volume ratio for both-handed strangulations (p = 0.011). Measurement of side differences in cervical muscle volume does not allow for a conclusion concerning the probable handedness of the perpetrator.

Towards a New Qualitative Screening Assay for Synthetic Cannabinoids Using Metabolomics and Machine Learning.

BACKGROUND: Synthetic cannabinoids (SCs) are steadily emerging on the drug market. To remain competitive in clinical or forensic toxicology, new screening strategies including high-resolution mass spectrometry (HRMS) are required. Machine learning algorithms can detect and learn chemical signatures in complex datasets and use them as a proxy to predict new samples. We propose a new screening tool based on a SC-specific change of the metabolome and a machine learning algorithm. METHODS: Authentic human urine samples (n = 474), positive or negative for SCs, were used. These samples were measured with an untargeted metabolomics liquid chromatography (LC)-quadrupole time-of-flight-HRMS method. Progenesis QI software was used to preprocess the raw data. Following feature engineering, a random forest (RF) model was optimized in R using a 10-fold cross-validation method and a training set (n = 369). The performance of the model was assessed with a test (n = 50) and a verification (n = 55) set. RESULTS: During RF optimization, 49 features, 200 trees, and 7 variables at each branching node were determined as most predictive. The optimized model accuracy, clinical sensitivity, clinical specificity, positive predictive value, and negative predictive value were 88.1%, 83.0%, 92.7%, 91.3%, and 85.6%, respectively. The test set was predicted with an accuracy of 88.0%, and the verification set provided evidence that the model was able to detect cannabinoid-specific changes in the metabolome. CONCLUSIONS: An RF approach combined with metabolomics enables a novel screening strategy for responding effectively to the challenge of new SCs. Biomarkers identified by this approach may also be integrated in routine screening methods.

RiFNet: Automated rib fracture detection in postmortem computed tomography.

Imaging techniques are widely used for medical diagnostics. In some cases, a lack of medical practitioners who can manually analyze the images can lead to a bottleneck. Consequently, we developed a custom-made convolutional neural network (RiFNet = Rib Fracture Network) that can detect rib fractures in postmortem computed tomography. In a retrospective cohort study, we retrieved PMCT data from 195 postmortem cases with rib fractures from July 2017 to April 2018 from our database. The computed tomography data were prepared using a plugin in the commercial imaging software Syngo.via whereby the rib cage was unfolded on a single-in-plane image reformation. Out of the 195 cases, a total of 585 images were extracted and divided into two groups labeled "with" and "without" fractures. These two groups were subsequently divided into training, validation, and test datasets to assess the performance of RiFNet. In addition, we explored the possibility of applying transfer learning techniques on our dataset by choosing two independent noncommercial off-the-shelf convolutional neural network architectures (ResNet50 V2 and Inception V3) and compared the performances of those two with RiFNet. When using pre-trained convolutional neural networks, we achieved an F1 score of 0.64 with Inception V3 and an F1 score of 0.61 with ResNet50 V2. We obtained an average F1 score of 0.91 ± 0.04 with RiFNet. RiFNet is efficient in detecting rib fractures on postmortem computed tomography. Transfer learning techniques are not necessarily well adapted to make classifications in postmortem computed tomography.

Sperm hunting on optical microscope slides for forensic analysis with deep convolutional networks - a feasibility study.

Microscopic sperm detection is an important task in sexual assault cases. In some instances, the samples contain no or only low amounts of semen. Therefore, the biological material is transferred onto a glass slide and needs to be manually scanned using an optical microscope. This work can be very time consuming, especially when no spermatozoa is present. In such a case, the result needs to be validated. In this article we show how convolutional neural networks can perform this task and how they can reduce the scanning time by locating the sperm cells on images taken under the microscope. For this purpose, we trained a VGG19 network and a VGG19 variation with 1942 images, some containing sperm cells and some not.

The forensic holodeck - Recommendations after 8 years of experience for additional equipment to document VR applications.

The forensic holodeck was first introduced in 2013, using the first upcoming commercially available virtual reality gaming headsets to visualize forensic 3D reconstructions. Following the publication of this development virtual reality was introduced in case work in a variety of different ways. After 8 years of using virtual reality in a professional forensic capacity this professional practice report will show, which equipment is necessary in addition to a virtual reality setup. This mostly includes audio-visual and broadcasting technology for complete documentation of the application of virtual reality, but also some other IT equipment, which should be available for as low as 20'000 US$. Guidelines, hints and tips regarding equipment acquisition, setup and use will be provided and discussed.

Interpretable machine learning model to detect chemically adulterated urine samples analyzed by high resolution mass spectrometry.

OBJECTIVES: Urine sample manipulation including substitution, dilution, and chemical adulteration is a continuing challenge for workplace drug testing, abstinence control, and doping control laboratories. The simultaneous detection of sample manipulation and prohibited drugs within one single analytical measurement would be highly advantageous. Machine learning algorithms are able to learn from existing datasets and predict outcomes of new data, which are unknown to the model. METHODS: Authentic human urine samples were treated with pyridinium chlorochromate, potassium nitrite, hydrogen peroxide, iodine, sodium hypochlorite, and water as control. In total, 702 samples, measured with liquid chromatography coupled to quadrupole time-of-flight mass spectrometry, were used. After retention time alignment within Progenesis QI, an artificial neural network was trained with 500 samples, each featuring 33,448 values. The feature importance was analyzed with the local interpretable model-agnostic explanations approach. RESULTS: Following 10-fold cross-validation, the mean sensitivity, specificity, positive predictive value, and negative predictive value was 88.9, 92.0, 91.9, and 89.2%, respectively. A diverse test set (n=202) containing treated and untreated urine samples could be correctly classified with an accuracy of 95.4%. In addition, 14 important features and four potential biomarkers were extracted. CONCLUSIONS: With interpretable retention time aligned liquid chromatography high-resolution mass spectrometry data, a reliable machine learning model could be established that rapidly uncovers chemical urine manipulation. The incorporation of our model into routine clinical or forensic analysis allows simultaneous LC-MS analysis and sample integrity testing in one run, thus revolutionizing this field of drug testing.

Correlation of age, sex and season with the state of human decomposition as quantified by postmortem computed tomography.

Dead bodies exhibit a variable range of changes with advancing decomposition. To quantify intracorporeal gas, the radiological alteration index (RAI) has been implemented in the assessment of postmortem whole-body computed tomography. We used this RAI as a proxy for the state of decomposition. This study aimed to (I) investigate the correlation between the state of decomposition and the season in which the body was discovered; and (II) evaluate the correlations between sociodemographic factors (age, sex) and the state of decomposition, by using the RAI as a proxy for the extent of decomposition. In a retrospective study, we analyzed demographic data from all autopsy reports from the Institute of Forensic Medicine of Zurich between January 2017 to July 2019 and evaluated the radiological alteration index from postmortem whole-body computed tomography for each case. The bodies of older males showed the highest RAI. Seasonal effects had no significant influence on the RAI in our urban study population with bodies mostly being discovered indoors. Autopsy reports contain valuable data that allow interpretation for reasons beyond forensic purposes, such as sociopolitical observations.

Potential use of deep learning techniques for postmortem imaging.

The use of postmortem computed tomography in forensic medicine, in addition to conventional autopsy, is now a standard procedure in several countries. However, the large number of cases, the large amount of data, and the lack of postmortem radiology experts have pushed researchers to develop solutions that are able to automate diagnosis by applying deep learning techniques to postmortem computed tomography images. While deep learning techniques require a good understanding of image analysis and mathematical optimization, the goal of this review was to provide to the community of postmortem radiology experts the key concepts needed to assess the potential of such techniques and how they could impact their work.

The cutting edge-an investigation into the pressure necessary for cutting skin with different knife blade types.

Sharp force trauma is routinely encountered in forensic practice. Often the question is posed, how much pressure or energy would have been necessary to inflict a cut with a specific knife, in order to further characterize the perpetrator or determine his intent to cause harm. This paper investigates two knife blades and its individual pressures needed to cut through the epidermis and dermis. In order to examine the necessary force for cutting through the skin, we performed experiments on a piglet skin-on-gelatin phantom. Two similarly small knives-a paring knife with a serrated blade and a Swiss Army pocket knife with a smooth blade-were moved over the phantom surface using a mobile cutting apparatus with varying weight on the load arm with the knife. The depth of the cut was to be determined according to a scale from zero to three: grade 0 = no cut; grade 1 = cut into epidermis only; grade 2 = cut into dermis; grade 3 = complete transection of the entire skin. Each cut inflicted at a specific pressure force was assessed closely in order to ascertain depth, calculate the velocity, and ultimately determine the point at which a cut would inflict grade 3 damage. The smooth blade of a pocket knife needed at least 1900g pressure in order to slice through pig skin mounted on a thick gelatin block, whereas a serrated blade of a paring knife managed to cut into or through the dermis at a comparatively lower force of 700g. Our study shows that at the same cutting velocity, a significant difference in pressure is necessary to inflict the same degree of damage.

A machine learning approach for handling big data produced by high resolution mass spectrometry after data independent acquisition of small molecules - Proof of concept study using an artificial neural network for sample classification.

Liquid chromatography coupled to high-resolution mass spectrometry (HRMS) enables data independent acquisition (DIA) and untargeted screening. However, to avoid the handling of the resulting large dataset, most laboratories in that field still use targeted screening methods, which offer good sensitivity and specificity but are limited to known compounds. The promising field of machine learning offers new possibilities such as artificial neural networks that can be trained to classify large amounts of data. In this proof of concept study, we exemplify such a machine learning approach for raw HRMS-DIA data files. We evaluated a machine learning model using training, validation, and test sets of solvent and whole blood samples containing drugs (of abuse) common in forensic toxicology. For that purpose, different platforms were used. With a feedforward neural network model architecture, a category prediction (blank sample vs. drug containing sample) was aimed for. With the applied machine learning approaches, the sensitivity and specificity, of the validation and test set, for the prediction of sample classes were in a suitable range for an actual use in a (routine) laboratory (e.g. workplace drug testing). In conclusion, this proof of concept study clearly demonstrated the huge potential of machine learning in the analysis of HRMS-DIA data.

A toolbox for the rapid prototyping of crime scene reconstructions in virtual reality.

Virtual reality is recently finding its way in forensic work. The required 3D data is nowadays a standard dataset available in many cases, from homicide to traffic collisions, including not only data from the scene but also of weaponry and involved persons. Current investigations use these 3D data to replicated the incident and as discussion base for forensic personal. However, modifying the scene on a 2D viewport is often cumbersome due to the loss of the third dimension. Also to perform the modifications on the scene a 3D operator is often required. Virtual reality might improve this step by its easy use and by visualising the third dimension. This publication presents a variety of tools which can be used in forensic investigations. Additionally to the tools, examples of forensic use of these tools will be presented, showing that already a small number of tools support a variety of forensic applications.

Microbiome-based body fluid identification of samples exposed to indoor conditions.

In the forensic reconstruction of crime scene activities, the identification of biological traces and their bodily origin are valuable evidence that can be presented in court. While several presumptive and confirmatory tests are currently available, the limitations in specificity and sensitivity have instigated a search for alternative methods. Bacterial markers have been proposed as a novel approach for forensic body fluid/tissue identification. Bacteria are not only ubiquitous throughout the human body, but also, as shown by recent microbiome sequencing studies of the 16S rRNA gene, bacterial community structures are distinct across body sites. Traces and stains at crime scenes are, however, often exposed to the environment outside the human body for variable periods of time before laboratory processing. Thus, it is not clear whether exposed samples continue to harbor microbial signatures characteristic of their body site of origin. In this proof-of-concept study we collected samples from six different body sites: saliva, skin, peripheral blood, vaginal fluid, menstrual blood and semen. We exposed a subset of these samples to indoor conditions for 30 days while the remaining samples were processed directly after extraction. Our analyses of 16S rRNA gene sequence data for a total of 46 control and exposed samples show that both types of samples group by body site, although a few outliers are observed. Based on our results, vaginal and menstrual samples share their microbial signatures, and cannot be distinguished using bacterial markers. Overall, our findings indicate that bacterial markers are a promising avenue for forensic body fluid/tissue identification.

Understanding policing as a mechanism of cheater control in cooperating bacteria.

Policing occurs in insect, animal and human societies, where it evolved as a mechanism maintaining cooperation. Recently, it has been suggested that policing might even be relevant in enforcing cooperation in much simpler organisms such as bacteria. Here, we used individual-based modelling to develop an evolutionary concept for policing in bacteria and identify the conditions under which it can be adaptive. We modelled interactions between cooperators, producing a beneficial public good, cheaters, exploiting the public good without contributing to it, and public good-producing policers that secrete a toxin to selectively target cheaters. We found that toxin-mediated policing is favoured when (a) toxins are potent and durable, (b) toxins are cheap to produce, (c) cell and public good diffusion is intermediate, and (d) toxins diffuse farther than the public good. Although our simulations identify the parameter space where toxin-mediated policing can evolve, we further found that policing decays when the genetic linkage between public good and toxin production breaks. This is because policing is itself a public good, offering protection to toxin-resistant mutants that still produce public goods, yet no longer invest in toxins. Our work thus highlights that not only specific environmental conditions are required for toxin-mediated policing to evolve, but also strong genetic linkage between the expression of public goods, toxins and toxin resistance is essential for this mechanism to remain evolutionarily stable in the long run.

Applying virtual reality in forensics - a virtual scene walkthrough.

A major task of forensic investigations is the documentation and interpretation of evidence to reconstruct a forensically relevant incident. To accomplish this task, a scene is documented not only with photographs but also with 3D documentation technologies. The resulting 3D data are used for 3D visualization and to perform 3D reconstructions. In this article, we present an approach for using forensic 3D data in conjunction with virtual reality to perform scene walkthroughs in the context of witness or suspect interrogations. The aim is to provide a method for scene visits showing the original scene even years after the incident. These scene walkthroughs in VR can be reproduced and allow to see through the eyes of a witness by recording their behavior and actions. These recordings allow subsequent examinations and reconstruction to support the investigation and scene understanding and can be used as evidence in court.

Female nursing partner choice in a population of wild house mice (Mus musculus domesticus).

BACKGROUND: Communal nursing in house mice is an example of cooperation where females pool litters in the same nest and indiscriminately nurse own and other offspring despite potential exploitation. The direct fitness benefits associated with communal nursing shown in laboratory studies suggest it to be a selected component of female house mice reproductive behaviour. However, past studies on communal nursing in free-living populations have debated whether it is a consequence of sharing the same nest or an active choice. Here using data from a long-term study of free-living, wild house mice we investigated individual nursing decisions and determined what factors influenced a female's decision to nurse communally. RESULTS: Females chose to nurse solitarily more often than expected by chance, but the likelihood of nursing solitarily decreased when females had more partners available. While finding no influence of pairwise relatedness on partner choice, we observed that females shared their social environment with genetically similar individuals, suggesting a female's home area consisted of related females, possibly facilitating the evolution of cooperation. Within such a home area females were more likely to nest communally when the general relatedness of her available options was relatively high. Females formed communal nests with females that were familiar through previous associations and had young pups of usually less than 5 days old. CONCLUSIONS: Our findings suggest that communal nursing was not a by-product of sharing the same nesting sites, but females choose communal nursing partners from a group of genetically similar females, and ultimately the decision may then depend on the pool of options available. Social partner choice proved to be an integrated part of cooperation among females, and might allow females to reduce the conflict over number of offspring in a communal nest and milk investment towards own and other offspring. We suggest that social partner choice may be a general mechanism to stabilize costly cooperation.

Transition from one- to two-dimensional development facilitates maintenance of multicellularity.

Filamentous organisms represent an example where incomplete separation after cell division underlies the development of multicellular formations. With a view to understanding the evolution of more complex multicellular structures, we explore the transition of multicellular growth from one to two dimensions. We develop a computational model to simulate multicellular development in populations where cells exhibit density-dependent division and death rates. In both the one- and two-dimensional contexts, multicellular formations go through a developmental cycle of growth and subsequent decay. However, the model shows that a transition to a higher dimension increases the size of multicellular formations and facilitates the maintenance of large cell clusters for significantly longer periods of time. We further show that the turnover rate for cell division and death scales with the number of iterations required to reach the stationary multicellular size at equilibrium. Although size and life cycles of multicellular organisms are affected by other environmental and genetic factors, the model presented here evaluates the extent to which the transition of multicellular growth from one to two dimensions contributes to the maintenance of multicellular structures during development.