Postmortem Computed Tomography Images Research Articles

Performance of postmortem CT in the diagnosis of natural death from out-of-hospital cardiac arrest.

Postmortem CT (PMCT) is used widely to identify the cause of death. However, its diagnostic performance in cases of natural death from out-of-hospital cardiac arrest (OHCA) may be unsatisfactory because the cause tends to be cardiogenic and cannot be detected on PMCT images. We retrospectively investigated the diagnostic performance of PMCT in the diagnosis of natural death from OHCA and compared it to that of unnatural death. Our series included 450 cases; 336 were natural- and 114 were unnatural death cases. Between 2018 and 2022 all underwent non-contrast PMCT to identify the cause of death. Two radiologists reviewed the PMCT images and categorized them as diagnostic (PMCT alone sufficient to determine the cause of death), suggestive (the cause of death was suggested but additional information was needed), and non-diagnostic (the cause of death could not be determined on PMCT images). The diagnostic performance of PMCT was defined by the percentage of diagnosable and suggestive cases and compared between natural- and unnatural death cases. Interobserver agreement for the cause of death on PMCT images was also assessed with the Cohen kappa coefficient of concordance. The diagnostic performance of PMCT for the cause of natural- and unnatural deaths from OHCA was 30.3% and 66.6%, respectively (p < 0.01). The interobserver agreement for the cause of natural- and unnatural deaths on PMCT images was very good with kappa value 0.92 and 0.96, respectively. As PMCT identified the cause of natural death by OHCA in only 30% of cases, its diagnostic performance must be improved.

Forensic significance and inferential value of PMCT features in charred bodies: A bicentric study

Post-mortem computed tomography (PMCT) is a diagnostic tool that provides essential information in forensic field. This radiological technique can play a pivotal role in personal identification, gunshot injuries, major traumatic injuries, child abuse, drowning asphyxia, embolism, COVID-19, sudden death, and charring cases.Post-mortem imaging can be used to investigate charred bodies as it allows the forensic examiner to distinguish traumatic and heat injuries. PMCT images can also provide additional information to determine whether the individual was alive at the time the fire started.In our bicentric study, we performed PMCT on 15 cases of charred bodies (10 males and 5 females) collected at the Florence University (Florence, Italy) and Catholic University of The Sacred Heart (Rome, Italy) before the external examination and/or full autopsy examination. In both the institutions, PMCT was performed through a Siemens SOMATOM 16-slice CT scanner (Siemens Aktiengesellschaft, Berlin, 2010).The aim of the current study was to obtain unambiguous information that could help determine the cause of death by comparing our PMCT findings with the most recent evidence in the scientific literature. Our investigations focused on the evaluation of bone texture and differences from traumatic fractures, leading also to an improvement in the interpretation of some aspecific findings (e.g., soot) from the integration of the radiological information with the autopsy results.The data obtained in our study support PMCT as a reliable diagnostic tool in the study of charred corpses and represent basic evidence for further applications of radiology in the forensic field.

The cardiothoracic ratio on postmortem CT as a screening tool for cardiomegaly in the Thai population.

IntroductionIn forensic radiology, the cardiothoracic ratio (CTR) calculated from postmortem computed tomography (PMCT) images can be used to detect cardiomegaly. In this study, a new measurement method is studied that involves measurement in the anteroposterior (AP) and transverse directions, with a reference level based on the Thai population. ObjectiveTo screen for cardiomegaly using the CTR calculated from PMCT images. Research methodA sample size of 116 deceased Thai individuals who underwent PMCT before autopsy was obtained. Individuals were divided into two groups: normal heart weight and overweight heart. Hearts heavier than the mean plus one standard deviation were categorized into the overweight group. The CTR was calculated in both the AP and transverse directions at six reference levels. Receiver operating characteristic curves (ROC) were calculated to determine the CTR cutoff point for the diagnosis of cardiomegaly. ResultsThe CTR cutoff values for diagnosing cardiomegaly were as follows: 1) CTR >0.5 in the transverse direction at the mid-vertebra of T7 (sensitivity, 75.6%; specificity, 70.6%; area under the ROC curve 0.81), 2) CTR >0.49 in the transverse direction at the mid-vertebra of T8 (sensitivity 71%, specificity 81.4%, area under the ROC curve 0.80). The selection of the cutoff values depended on the location of the heart. The area under the ROC curve in the AP direction was in the range 0.5-0.7, which is inferior compared to the transverse direction (0.7-0.9). ConclusionCalculating the CTR with a reference level on PMCT images can assist in the diagnosis of cardiomegaly.

U-Net-Based Automatic Segmentation of Sphenoid Sinus Fluid in Drowning Cases Using Postmortem CT Images: A Feasibility Study

Detecting sphenoid sinus fluid (SSF) is an additional finding in autopsies for diagnosing drowning. SSF can provide additional forensic evidence through laboratory tests such as diatom and electrolyte analyses. If drowning is suspected, accurately assessing the presence and volume of SSF during an autopsy is crucial. Utilizing postmortem computed tomography (PMCT) images could aid in accurately sampling SSF. Accurately segmenting the region of interest is essential for volume analysis using computed tomography images. However, manual segmentation techniques are labor-intensive and time-consuming, and their success depends on the experience of the observer. Therefore, this study aimed to develop a U-Net–based deep learning model for the automatic segmentation of SSF in drowning cases using PMCT images and to evaluate the performance of the model. We retrospectively reviewed 34 drowning cases in which both PMCT scans and forensic autopsies were performed at our institution. The U-Net architecture of deep learning was used for automatic segmentation. The proposed model achieved the Dice similarity coefficient (DSC) and Intersection over Union (IoU) of a maximum of 95.85% and 92.03%, a minimum of 0% and 0%, and an average of 77.15% and 67.18%, respectively. Although the average DSC and IoU did not show high similarity, this study showed that PMCT images can be used for automatic segmentation of SSF in drowning cases, which could improve the performance with sufficient dataset acquisition and further model training.

Distinguishing true from pseudo hematoma in the cervical spinal canal using post-mortem computed tomography

Spinal cord injury is difficult to detect directly on postmortem computed tomography (PMCT) and it is usually diagnosed by indirect findings such as a hematoma in the spinal canal. However, we have encountered cases where the hematoma-like high-attenuation area in the cervical spinal canal was visible on PMCT, while no hematoma was observed at autopsy; we called it a “pseudo hematoma in the cervical spinal canal (pseudo-HCSC).” In this retrospective study, we performed statistical analysis to distinguish true from pseudo-HCSC. The cervical spinal canal was dissected in 35 autopsy cases with a hematoma-like high-attenuation area (CT values 60-100 Hounsfield Unit (HU)) in the spinal canal from the first to the fourth cervical vertebrae in axial slices of PMCT images. Of these 22 had a hematoma and 13 did not (pseudo-HCSC). The location and length of the hematoma-like high-attenuation and spinal cord areas were assessed on reconstructed PMCT images, true HCSC cases had longer the posterior hematoma-like area and shorter the spinal cord area in the midline of the spinal canal (P < 0.05). Furthermore, we found that true HCSC cases were more likely to have fractures and gases on PMCT while pseudo-HCSC cases were more likely to have significant facial congestion (P < 0.05). We suggest that pseudo-HCSC on PMCT is related to congestion of the internal vertebral venous plexus. This study raises awareness about the importance of distinguishing true HCSC from pseudo-HCSC in PMCT diagnosis, and it also presents methods for differentiation between these two groups.

Pericardial Effusion Detection on Post-Mortem Computed Tomography Images Using Convolutional Neural Networks.

Pericardial effusion can be a sign of significant underlying diease and, in some cases, may lead to death. Post-mortem computed tomography (PMCT) is a well-established tool to assist death investigation processes in the forensic setting. In practice, the scarcity of well-trained radiologists is a challenge in processing raw whole-body PMCT images for pericardial effusion detection. In this work, we propose a Pericardial Effusion Automatic Detection (PEAD) framework to automatically process raw whole-body PMCT images to filter out the irrelevant images with heart organ absent and focus on pericardial effusion detection. In PEAD, the standard convolutional neural network architectures of VGG and ResNet are carefully modified to fit the specific characteristics of PMCT images. The experimental results prove the effectiveness of the proposed framework and modified models. The modified VGG and ResNet models achieved superior detection accuracy than the standard architecture with reduced processing speed.

Forensic age estimation of the knee by post-mortem DR, CT, and MR imaging: a comparative study

It is believed by many that reference data for age estimation purposes must be imaging-modality specific. A study from our department has however proven otherwise. We therefore found it interesting to investigate this further by looking at the level of agreement between different imaging modalities. The aim of this study was to investigate the level of agreement between the three radiological modalities, computed tomography (CT), magnetic resonance imaging (MRI), and digital radiography (DR), in assessing the ossification of the epiphyses of the knee. A total of 34 deceased individuals of 10–25 years of age, brought in for a medicolegal autopsy at our department, were scanned by CT, MRI, and DR. The ossification stages of the three bones of the right knee, distal femoral, proximal tibial, and proximal fibular epiphysis were assessed using the established combined staging method by Schmeling et al. and Kellinghaus et al. Analysis of the results by Cohen’s weighted kappa showed a good agreement between CT and DR (K = 0.61–0.70), and MRI and DR (K = 0.68–0.79) but only moderate agreement between CT and MRI (K = 0.55–0.57). This leads us to conclude that different radiological images cannot be used interchangeably for age estimation purposes, so reference material needs to be imaging-modality specific. However, to make a more general conclusion research on a larger population is needed.

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.

Estimation of missing mandible from the skull using postmortem CT images

Skeletal remains often have missing mandibles owing to the environment in which the corpse was placed or damage caused by an animal. Loss of the mandible reduces the accuracy of skull identification. Although several studies have validated methods for estimating mandibular morphology from conventional anthropological measurements using skull specimens, there are no reports using three-dimensional computed tomography (3DCT) images. Here, we examined methods for establishing the mandibular morphology from the remaining skulls using postmortem computed tomography (PMCT) images. We used PMCT images from 200 Japanese subjects as samples. After verifying the morphological correlation between the cranium and mandible, we created and validated estimation models using multiple regression analysis (stepwise method) for seven sites that were necessary for understanding the morphology of the mandible. Among the regression models, the estimated model for bicondylar breadth had the highest coefficient of determination (adjR2 = 0.53). We verified the accuracy of the model on a sample independent from the specimen used to create the estimation model and found that the formulated model of bicondylar breadth had good estimation accuracy, with a high correlation coefficient between the measured and predicted values of 0.82 and a mean absolute error of 3.582 mm, indicating that the model had good estimation accuracy. Here, we established a novel method for estimating the missing mandibular morphology based on PMCT data from the Japanese population. Our estimation model can help determine the missing mandibular morphology in a cranium remnant.

Objective Evaluation of Oral and Pharyngeal Areas in Autopsy Cases of Obstructive Sleep Apnea Syndrome via Postmortem CT.

Obstructive sleep apnea syndrome (OSAS) can cause sudden death during sleep. Previous findings have suggested that OSAS development is related to maxillofacial morphology. Evaluation of facial morphology can determine the risk of developing the disease, and establishing an objective method to assess the underlying etiology of OSAS-related death would be advantageous. The objective of this study is to determine the key features of obstructive sleep apnea syndrome (OSAS) using postmortem oral and pharyngeal computed tomography (CT). We retrospectively assessed autopsy cases of patients with (n=25) and without (n=25) OSAS-related death. We used oral and pharyngeal CT images to compare the oral and pharyngeal cavity volume (OPCV), oral and pharyngeal soft tissue volume (OPSV), oral and pharyngeal air space volume (OPAV), and OPAV to OPCV ratio (%air). Receiver operating curve (ROC) analysis was used to determine the accuracy of OSAS prediction. We assessed participants with body mass index (BMI) values within the normal range. Among the 50 subjects, we observed significant between-group differences in OPSV, OPAV, and % air, whereas there were significant between-group differences in OPSV and %air among 28 subjects with normal BMI values. Both comparisons suggested that OSAS-related death was associated with low %air and high OPSV values. The %air and OPSV are useful for assessing postmortem oropharyngeal CT images. OSAS-related sudden death is likely when %air and OPSV values are ≤20.1% and ≥127.2 ml, respectively. Among those with normal BMI values, % air and OPSV values of ≤22.8% and ≥111.5 ml, respectively, predict OSAS-related sudden death.

Simplified Formula for Estimating Nasal Dimensions for 3-Dimensional Facial Reconstruction among Japanese Adults

The eyelids, external nose, and lips play an important role in individual identification and facial recognition; however, they are excluded from tissue marker points, and are reconstructed based on generic methods for 3D facial reconstruction or facial approximation. Therefore, this study focused on nasal dimensions and evaluated whether Krogman’s widely used formula for estimating the dimensions of an external nose, regardless of sex, race, and body physique, can be applied to Japanese adults. A total of 146 postmortem CT images of Japanese adult cadavers (64 males and 82 females, aged 58–105 years old) were retrospectively analyzed. The total nasal projection (TNP) among Japanese adults was estimated using the formula, TNP = 1.9 × the anterior nasal spine projection (ANSP) + the mid-philtrum depth (MPD), which differed significantly from the coefficient (3.0) in the conventional formula, regardless of sex, race, and body physique, and therefore needed modification for Asians. Although there was no positive relationship between the total nasal width (TNW) and the maximum width of the anterior nasal aperture (ANAW), the TNW could be estimated by adding soft tissue that varies by sex and body physique to both sides of the nearly constant ANAW. Therefore, we determined a simple and practical formula to estimate nasal dimensions among Japanese adults for conventional 3D facial reconstruction and manual 3D facial sculpture.

Post-mortem pulmonary tuberculosis: comparison of available diagnostic methods.

Tuberculosis (TB) caused by Mycobacterium tuberculosis remains a serious public health threat with the World Health Organisation (WHO) reporting 5.8 million cases and 1.3 million deaths in the year 2020 due to TB. TB can be diagnosed by imaging, histopathological and bacteriological methods with culture remaining the gold standard. This study was performed to look at the sensitivity and specificity of post-mortem computed tomography (PMCT) imaging when compared to culture in diagnosing pulmonary tuberculosis. This was a retrospective comparative study looking at post mortem cases where lung tissue samples sent for TB culture at Hospital Kuala Lumpur were compared against PMCT imaging. Exclusion criteria included contaminated samples, decomposed cases, immunocompromised subjects and those below 18 years of age. Subjects included 80 medico-legal autopsy cases at the National Institute of Forensic Medicine, Hospital Kuala Lumpur, Malaysia who had whole body PMCT done in accordance with the Institute's protocol and tissue samples sent for bacteriology culture for tuberculosis. PMCT findings were positively associated with acid-fast organisms in 23.5 out of 33 cases (71.2%). Our study also showed that PMCT had a sensitivity of 71.3% and specificity of 54.3% (95% CI: 39.5-68.4) in diagnosing TB based on the protocol set in this study. This study showed that there was relatively good agreement between radiological PMCT findings and bacterial culture, suggesting that radiological examination is a relatively reliable tool for preliminary screening and possible diagnosis of TB prior to a postmortem examination which would be beneficial in reducing the risk of transmission of TB to health workers during autopsy.

Acute subdural hematoma caused by rupture of a mycotic aneurysm due to meningitis associated with infectious endocarditis: comparison of autopsy findings with postmortem computed tomography.

Forensic pathologists often encounter cases of acute subdural hematoma (SDH) due to trauma, whereas those attributable to endogenous causes are rare. Here, we report a case of the latter type in a 42-year-old man who was found dead at home after several months of fever and malaise. Postmortem computed tomography (PMCT) and autopsy were undertaken to clarify the cause of death. PMCT images revealed a fatal SDH and a localized hyper-density area in the right parietal lobe; macroscopic and microscopic examinations revealed SDH due to rupture of a mycotic aneurysm (MA) associated with meningitis. The PMCT images also indicated thickening and calcification of the mitral valve, while autopsy demonstrated infective endocarditis (IE). In addition, PMCT demonstrated a low-density area in the spleen, which was shown to be a splenic abscess at autopsy. PMCT also demonstrated tooth cavities. Based on the findings of autopsy, the cause of death was considered to be SDH due to rupture of the MA resulting from meningitis with IE and splenic abscess. Although PMCT was unable to clarify the significance of any individual feature, a retrospective review of the PMCT images might have suggested IE, bacteremia, or ruptured MA leading to SDH. This case suggests that, instead of interpreting individual features demonstrated on PMCT images, integrated interpretation of overall PMCT findings might provide clues for identifying causes of death, despite the fact that PMCT lacks diagnostic accuracy for infectious diseases such as IE and meningitis.

Examination of postmortem changes in the lungs, trachea, and bronchi in a rat model imaged with small-animal computed tomography.

As less autopsies are performed, the need for postmortem computed tomography (PMCT) as an alternative is increasing. It is important to know how postmortem changes over time are reflected on CT, in order to improve the diagnostic capability of PMCT and replace forensic pathology evaluations such as time of death estimation. In this study, we examined temporal changes on postmortem chest CT images of a rat model. After acquiring antemortem images under isoflurane inhalation anesthesia, the rats were euthanized with a rapid intravenous injection of anesthetics. From immediately after death to 48 hours postmortem, chest images were acquired using small-animal CT. The 3D images were then evaluated on a workstation to measure the antemortem and postmortem air content in the lungs, trachea, and bronchi over time. The air content in the lungs decreased, but the air content of the trachea and bronchi temporarily increased 1-12 hours postmortem, then decreased at 48 hours postmortem. Therefore, the measurement of trachea and bronchi volumes on PMCT could be an objective way to estimate the time of death. While the air content of the lungs decreased, the volume of the trachea and bronchi temporarily increased after death, indicating the potential to use such measurements to estimate time of death.

Estimating age at death by Hausdorff distance analyses of the fourth lumbar vertebral bodies using 3D postmortem CT images.

The existing methods for determining adult age from human skeletons are mostly qualitative. However, a shift in quantifying age-related skeletal morphology on a quantitative scale is emerging. This study describes an intuitive variable extraction technique and quantifies skeletal morphology in continuous data to understand their aging pattern. A total of 200 postmortem CT images from the deceased aged 25-99years (130 males, 70 females) who underwent forensic death investigations were used in the study. The 3D volume of the fourth lumbar vertebral body was segmented, smoothed, and post-processed using the open-source software ITK-SNAP and MeshLab, respectively. To measure the extent of 3D shape deformity due to aging, the Hausdorff distance (HD) analysis was performed. In our context, the maximum Hausdorff distance (maxHD) was chosen as a metric, which was subsequently studied for its correlation with age at death. A strong statistically significant positive correlation (P < 0.001) between maxHD and age at death was observed in both sexes (Spearman's rho = 0.742, male; Spearman's rho = 0.729, female). In simple linear regression analyses, the regression equations obtained yielded the standard error of estimates of 12.5years and 13.1years for males and females, respectively. Our study demonstrated that age-related vertebral morphology could be described using the HD method. Moreover, it encourages further studies with larger sample sizes and on other population backgrounds to validate the methodology.

A novel method to estimate adult age from the lumbar vertebral body using 3D PMCT images in Japanese

This study evaluated the age-related changes in the vertebral body using 3D Postmortem CT (PMCT) images and proposed an alternative age estimation formula. The PMCT images of 200 deceased individuals aged 25 to 99 years (126 males, 74 females) were retrospectively reviewed and included in the study. Using the open-source software ITK-SNAP and MeshLab, a 3D surface mesh of the fourth lumbar vertebral body (L4) and its convex hull models were created from the PMCT data. Using their inbuilt tools, volumes (in mm3) of the L4 surface mesh and convex hull models were subsequently computed. We derived VD, defined as the difference in volumes between the convex hull and L4 surface mesh normalized by L4 mesh volume, and VR, defined as the ratio of L4 mesh volume to convex hull volume based on individual L4. Correlation and regression analyses were performed between VD, VR, and chronological age. A statistically significant positive correlation (P < 0.001) between chronological age and VD, (rs = 0.764, males; rs = 0.725, females), and a significant negative correlation between chronological age and VR (rs = -0.764, males; rs = -0.725, females) was obtained in both sexes. The lowest standard error of the estimate was demonstrated by the VR at 11.9 years and 12.5 years for males and females, respectively. As such, their regression models to estimate adult age were Age = 248.9–2.5VR years, males; Age = 258.1–2.5VR years, females. These regression equations may be useful for estimating age in Japanese adults in forensic settings.

Implementation of a personal identification system using alveolar bone images

ObjectiveIn recent years, personal identification has been performed using antemortem panoramic X-ray images and postmortem-CT images. Using these, we have developed a personal identification method that focuses on the alveolar bone. This study examined the effectiveness of this method and aimed to implement a reproducible system. Materials and methodsFor personal identification, a total of 633 CT images and panoramic X-ray images belonging to three groups with different conditions were used.　These images were 160 sets in the same person group and 96,820 in the other groups. The similarity of alveolar bone images was calculated using the landmark method of Procrustes analysis. The processes were system implemented and the methodology was validated. ResultsThe ability to identify between the same person group and other person groups showed 0.9769 as the area under the curve (AUC: ROC curve). At the cutoff value of 4.978, there was no false rejection rate, but false acceptance rate was slightly higher. ConclusionThis method was useful as a screening method for personal identification. In addition, system implementation was efficient and reduced human error. In the future, we aim to realize a more efficient personal identification method using distortion-corrected images and including auto-detective landmarks using deep learning.

Aortic calcification: A postmortem CT validation study in a middle-aged population

BackgroundComputed tomography (CT)-detected aortic calcification is strongly associated with aortic stiffness and is an accurate predictor of cardiovascular and all-cause mortality and cognitive decline. Some previous pathologic studies have shown calcium accumulation in the medial layer of the vessel wall, while others have suggested localisation in the atherosclerotic intimal layer. ObjectivesThe aim of this study was to histologically validate CT findings of aortic calcification for detectability and location in the aortic wall. MethodsWe acquired postmortem CT images and collected 170 aortic tissue samples from five different locations in the thoracic and abdominal aorta of 40 individuals who underwent autopsy. Microscopic slides were stained with haematoxylin and eosin and elastic van Gieson stain. Calcified lesions were characterised and calcifications were manually annotated in the intima and media. The presence and morphology of calcifications were scored on CT images. ResultsThe mean age of the autopsied individuals was 63 years, and 28 % died of cardiovascular disease. Calcifications were present in 74/170 (44 %) samples. Calcification was more common in the abdominal aorta than in the thoracic aorta. In all samples with calcifications, 99 % were located in the intimal layer. Only 16/170 samples had a small amount of medial arterial calcification. The histological results showed an 85 % concordance for the presence or absence of CT calcifications. There was complete inter-method agreement for annularity of calcifications in 68 % of the samples (linear weighted kappa 0.68 (95 %CI 0.60–0.77). ConclusionsAortic calcifications visible on CT are located in the intimal layer of the abdominal aorta wall, at least in aortas that are not aneurysmatic or dissected. The presence and annularity of these calcifications can be reliably determined by CT.

Diagnosing Drowning in Postmortem CT Images Using Artificial Intelligence.

Imaging features of the lung in postmortem computed tomography (CT) scans have been reported in drowning cases. However, it is difficult for forensic pathologists with limited experience to distinguish subtle differences in CT images. In this study, artificial intelligence (AI) with deep learning capability was used to diagnose drowning in postmortem CT images, and its performance was evaluated. The samples consisted of high-resolution CT images of the chest of 153 drowned and 160 non-drowned bodies captured by an 8- or 64-row multislice CT system. The images were captured with an image slice thickness of 1.0 mm and spacing of 30 mm, and 28 images were typically captured. A modified AlexNet was used as the AI architecture. The output result was the drowning probability for each component image. To evaluate the performance of the proposed model, the area under the receiver operating characteristic curve (AUC) was analyzed, and the AUC value of 0.95 was obtained. This indicates that the proposed AI architecture is a useful and powerful complementary testing approach for diagnosing drowning in postmortem CT images. Notably, the accuracy was 81% (62/77) for cases in which resuscitation was performed, and 92% (216/236) for cases in which resuscitation was not attempted. Therefore, the proposed AI method should not be used to diagnose the cause of death when aggressive cardiopulmonary resuscitation was performed. Additionally, because honeycomb lungs are likely to exhibit different morphologies, emphysema cases should also be treated with caution when the proposed AI method is used to diagnose drowning.

Deep Learning-Based Diagnosis of Fatal Hypothermia Using Post-Mortem Computed Tomography.

In forensic medicine, fatal hypothermia diagnosis is not always easy because findings are not specific, especially if traumatized. Post-mortem computed tomography (PMCT) is a useful adjunct to the cause-of-death diagnosis and some qualitative image character analysis, such as diffuse hyperaeration with decreased vascularity or pulmonary emphysema, have also been utilized for fatal hypothermia. However, it is challenging for inexperienced forensic pathologists to recognize the subtle differences of fatal hypothermia in PMCT images. In this study, we developed a deep learning-based diagnosis system for fatal hypothermia and explored the possibility of being an alternative diagnostic for forensic pathologists. An in-house dataset of forensic autopsy proven samples was used for the development and performance evaluation of the deep learning system. We used the area under the receiver operating characteristic curve (AUC) of the system for evaluation, and a human-expert comparable AUC value of 0.905, sensitivity of 0.948, and specificity of 0.741 were achieved. The experimental results clearly demonstrated the usefulness and feasibility of the deep learning system for fatal hypothermia diagnosis.