Advances in sudden unexplained death: novel insights into technical innovations and their applications in forensic investigation

Cardiac channelopathy

Among all cardiovascular diseases, deleterious genetic alterations in cardiac ion channel proteins are characterized by the absence of distinct pathological alterations, yet they can give rise to life-threatening arrhythmias that often present as SCD. Hereditary primary electrical disorders, including long QT syndrome (LQTS), short QT syndrome (SQTS), Brugada syndrome (BrS), and catecholaminergic polymorphic ventricular tachycardia (CPVT), are known as cardiac channelopathies and account for up to 30% of all SCDs in the young^[24], which are most commonly caused by mutations in genes encoding cardiac ion channels, receptors and/or their regulatory proteins^[15,25]. While specific clinical and ECG features can identify patients at the highest risk of SCD, stratifying those at lower risk remains challenging, as their first clinical manifestation may be sudden cardiac arrest or SCD^[26].

Genetic analyzing of deleterious genetic alterations of patients and their family members has become an important method for diagnosis of cardiac channelopathies in clinic^[27], and may be an optional method in forensic identification. However, due to large amount of unknown genes and rare variants associated with channelopathies, and many previously reported variants seem to have disputed evidence in pathogenesis of channelopathies, genetic analysis is still not a routine method in the clinic and forensic practice in China yet. Based on the current Chinese expert consensus statement on SCD, genetic testing is recommended for SCD cases in clinic with a suspected inherited etiology to identify causative variants, facilitate cascade screening of at-risk relatives, and enable accurate etiological diagnosis^[28]. In detail, for individuals with a defined cardiac phenotype, targeted gene panel testing focusing on well-established causative genes is preferred (Class I). In young (< 40 years) or phenotype-negative SCD cases with features suggestive of inherited arrhythmia syndromes (e.g., documented arrhythmic death or specific triggers), genetic evaluation focusing on primary arrhythmia syndrome genes, with optional inclusion of cardiomyopathy genes, is advised (Class I/IIa). Comprehensive postmortem sample collection (blood or fresh/fixed cardiac tissue) and multidisciplinary collaboration are emphasized to improve diagnostic yield, while the interpretation of variants of uncertain significance (VUS) requires caution and periodic re-evaluation^[28]. Correspondingly, forensic guidelines also need to be updated to include genetic testing in cases of SUD.

Concealed cardiomyopathy

In the past 5 years, the concept of “concealed cardiomyopathy (CCM)” has emerged to address a critical subset of the cases lacking identifiable cardiac structural abnormalities after comprehensive macroscopic and microscopic examination, while deleterious alteration in a gene associated with cardiomyopathy was identified in postmortem genetic testing^[29,30]. This entity highlights a crucial pathophysiological phase that malignant arrhythmias can occur prior to the development of overt structural changes in inherited cardiomyopathy^[31].

In autopsy-inconclusive cohorts, the prevalence of CCM is considerable. In a landmark study, 15.4% (14/91) of autopsy-inconclusive SCD cases were attributed to CCM, with 70% of clinically actionable variants found in cardiomyopathy-associated genes and nearly two-thirds involving arrhythmogenic cardiomyopathy (ACM) genes^[31]. The diagnostic yield was significantly higher in cases with subdiagnostic structural findings than in those with a completely normal heart. Postmortem genetic studies have detected pathogenic/likely pathogenic variants in up to 33% of young individuals diagnosed with myocarditis, suggesting inflammation can trigger lethal arrhythmias in genetically vulnerable myocardium^[32]. Beyond the decedent, identifying a pathogenic or likely pathogenic variant enables cascade testing, as shown by 42.6% of first-degree relatives testing positive for the familial variant, of whom 65% exhibited an observable phenotype, while 27 genotype-negative relatives were released from follow-up^[30,31].

In conclusion, CCM represents an under-recognized but clinically significant cause of SUDs, particularly in the young. Comprehensive evaluation should include not only a meticulous conventional autopsy but also broad, multi-phenotype genetic testing encompassing both ion channel and cardiomyopathy-associated genes, coupled with multidisciplinary family assessment^[29,33].

Epilepsy

SUD in epilepsy (SUDEP) is a sudden and unexpected, and non-traumatic death of a person with epilepsy, occurring without a witnessed seizure, and without a toxicological or pathological cause detected during the postmortem examination^[10,14]. SUDEP is a leading cause of mortality in epilepsy, accounting for up to 18% of deaths in patients with epilepsy^[34]. Main mechanisms have been proposed for SUDEP: (1) cardiac dysfunction, including arrhythmias and other cardiac events; (2) respiratory factors, such as seizure-induced pulmonary dysfunction; (3) cerebral and autonomic nervous system dysregulation; and (4) some less frequent contributors including anti-epileptic drugs, vagal nerve stimulation, and genetic factors, such as rare variants in genes associated with increased susceptibility to SUDEP^[35]. Seizure-induced apnea is a major event leading to death after generalized seizures^[36]. Dravet syndrome, developmental and epileptic encephalopathy (DEE) and focal epilepsy are closely related to increasing risk of SUDEP^[37].

The postmortem diagnosis of SUDEP is one of exclusion, relying primarily on witnessed seizures, medical records, and clinical history, and genetic testing has emerged as a highly effective tool in recent years. Over 33% of SUDEP cases are associated with deleterious genetic alterations in channelopathy-related genes, including some that are also implicated in SCD, such as variants in KCNH2 and SCN5A^[38,39]. These variants may increase susceptibility to seizure-induced cardiorespiratory dysfunction or fatal arrhythmias^[40]. Postmortem genetic findings allow for reclassification of the COD and enable cascade genetic screening of first-degree relatives, who may be at risk of inheritable cardiac channelopathies^[41]. Moreover, knowledge of a genetic risk factor may guide anti-seizure medication choices (e.g., avoiding sodium channel blockers in patients with SCN1A or SCN8A variants)^[42]. Therefore, genetic analysis should not be viewed solely as a research tool but rather integrated into SUDEP risk assessment and COD identification where available.

Sudden infant death syndrome

SIDS is now classified as SUD in the absence of both intrinsic and extrinsic risk factors^[43]. The incidence of SIDS varies among studies, accounting for 42% to 66% of all infant deaths^[44,45], and is typically higher between 2 and 4 months of age^[46]. SIDS occurs in infants with no identifiable intrinsic health risks and no modifiable extrinsic factors related to the sleeping environment, that is, the infant’s position and location^[47]. Given that deaths are often unobserved or unwitnessed, a standardized death-scene investigation is critical and recommended for establishing the COD to exclude any possibility of child abuse, unsafe products and environments^[48]. Our current understanding of SIDS is predominantly derived from epidemiological studies conducted in Western countries, particularly Europe and North America. In China, there is a paucity of reported SIDS cases, and neither clinical guidelines nor expert consensus statements on SIDS are currently available.

Increasing evidence has shown that SIDS shares common molecular and pathophysiological mechanisms with SADS and SUDEP, particularly involving cardiac arrhythmias^[14]. Other genetic background, such as rare variants in genes associated with inherited metabolic conditions, immune dysfunction and nicotine response, may also play a causal role in SIDS^[49], necessitating molecular autopsy including genetic analysis for elucidation of COD.

Postmortem decomposition

Postmortem decomposition, particularly putrefaction, poses substantial challenges in forensic pathology by progressively degrading tissues and obscuring critical morphological and biochemical evidence. The ability to determine an accurate COD is severely compromised in decomposed bodies, primarily due to the loss of histoarchitectural integrity, the introduction of postmortem artifacts, and the limitations imposed on conventional and advanced diagnostic techniques. In previous studies, putrefactive changes significantly impair the forensic evaluation of cardiac pathology. Critical diagnostic features are often lost: conventional histology of the myocardium is frequently unreliable due to autolysis and bacterial overgrowth, and myocardial wall thinning caused by putrefaction may mimic dilated cardiomyopathy, creating a substantial risk of misinterpretation^[50]. Radiological techniques are also hindered by decomposition artifacts, including gas accumulation, vascular collapse, and fluid shifts, all of which impair visualization of coronary arteries. Currently, no standardized forensic methodology exists to consistently differentiate antemortem cardiac pathology from postmortem artifacts^[51].

To define Post-Mortem Diagnostic Uncertainty, Van Den Bogaert et al. classify putrefaction as “lack of data due to post-mortem changes and destructive forces”, under which conditions conventional histological and toxicological analyses often fail. In cases of advanced decomposition, certain investigations become impractical or impossible, and unlike incomplete autopsies, such data loss is beyond the control of the forensic pathologist^[52]. Moreover, conditions without morphological abnormalities (e.g., channelopathies) remain a persistent source of uncertainty, as decomposition destroys even the limited structural evidence that might otherwise be present^[50].

In summary, putrefaction profoundly complicates COD determination by degrading diagnostic substrates and introducing artifacts. However, by recognizing both the persistent features (such as calcified plaques) and the limitations of standard techniques, forensic pathologists can still derive valuable information. Structured uncertainty communication and the adoption of validated ancillary methods are essential steps toward improving diagnostic accuracy in decomposed bodies^[52].

Sanger sequencing

Early applications of molecular autopsy predominantly relied on Sanger sequencing to analyze a limited number of pathogenic variants in key arrhythmia-related genes^[55], which remains the gold standard for DNA sequencing due to its high accuracy and reliability. According to current guidelines^[58], several genes associated with SCD can be effectively analyzed using Sanger sequencing, such as KCNQ1, KCNH2, SCN5A, RYR2, and LMNA^[10,58]. These genes are commonly implicated in inherited arrhythmia syndromes and cardiomyopathies and were the primary targets in early molecular autopsy studies. Moreover, Sanger sequencing also serves as the gold standard for targeted cascade genetic testing of at-risk first-degree relatives, enabling rapid and cost-effective determination of carrier status to guide clinical surveillance and preventive interventions^[59]. With the rapid expansion in the number of genes linked to SUD, Sanger sequencing is increasingly limited by its low throughput and inefficiency in analyzing large gene sets. Consequently, NGS technologies have progressively replaced Sanger sequencing as the primary discovery tool, owing to their high-throughput capacity and scalability^[53]. In current practice, Sanger sequencing is more commonly used as a confirmatory method for validating variants identified through NGS.

Next-generation sequencing

The emergence of NGS technologies has fundamentally transformed the analytical capacity of molecular autopsy, enabling high-throughput and cost-effective interrogation of genetic variation on an unprecedented scale^[53,54]. By allowing massively parallel sequencing of millions of DNA fragments, NGS has shifted postmortem genetic analysis from limited, hypothesis-driven testing toward more comprehensive and systematic approaches^[60]. NGS-based strategies can be implemented at multiple levels, including targeted gene panels, WES, and WGS, each offering distinct advantages depending on the clinical or forensic context.

Variant interpretation in molecular autopsy

Given the large number of variants identified in each individual, systematic filtering represents a critical first step. Technical filtering is applied to minimize false-positive calls arising from sequencing or alignment errors, commonly using thresholds such as a minimum read depth (e.g., ≥ 30×) and variant allele frequency (e.g., ≥ 20%). Subsequently, biological filtering is performed to distinguish rare, potentially relevant variants from background genetic noise. This process typically incorporates population frequency data, where variants with a minor allele frequency below 0.1% are prioritized, alongside consideration of gene relevance, particularly those previously associated with inherited arrhythmia syndromes or cardiomyopathies^[64].

Following filtering, variant prioritization aims to assess the potential functional and clinical significance of the remaining candidates^[65,66]. Public databases such as ClinVar and OMIM and in silico prediction tools including SIFT, PolyPhen-2, and MutationTaster are widely used to estimate the potential impact of variants on protein structure and function. Evolutionary conservation analysis further refines prioritization, as variants affecting highly conserved residues or domains are more likely to be functionally significant. To standardize variant interpretation, the American College of Medical Genetics and Genomics (ACMG), in collaboration with the Association for Molecular Pathology (AMP), has established a widely adopted five-tier classification system, categorizing variants as pathogenic, likely pathogenic, VUS, likely benign, or benign^[65]. This framework integrates multiple lines of evidence, including population data, computational predictions, functional assays, and segregation information, enabling a structured and transparent evaluation process. According to the ACMG/AMP guidelines, pathogenic and likely pathogenic variants are defined as deleterious, disease-causing alterations supported by sufficient evidence across multiple domains including population frequency, computational prediction, functional assays, and segregation data, whereas VUS indicates insufficient or conflicting evidence to determine their clinical effect and are considered non-actionable for clinical management or cascade testing^[67].

In the investigation of SUD, a stepwise genetic testing strategy is generally recommended^{[61,62,68,69]}. Targeted gene panels focusing on genes closely related to the suspected phenotype are often used as a first-line approach. While pathogenic/likely pathogenic variants provide actionable results that warrant cascade genetic screening in relatives, VUS findings represent a major interpretative dilemma because post-mortem tissues preclude functional validation, segregation analysis, and phenotypic correlation in the deceased proband^[70]. Given the large number of variants identified through NGS, variant interpretation has become a critical step, requiring careful filtering, prioritization, and classification within established frameworks. Importantly, the integration of molecular findings with clinical and familial information substantially enhances interpretative accuracy^[54,71]. For molecular autopsy specifically, international recommendations advocate that all rare variants identified in inherited arrhythmic syndrome genes be urgently reclassified following ACMG guidelines, with empirical studies showing that over 90% of previously unstandardized rare variants modified their classification upon rigorous ACMG application, and a maximum reanalysis interval of five years has been proposed to incorporate evolving evidence from population databases, case reports, and in silico tools^[53]. Ultimately, accurate interpretation of rare variants after molecular autopsy necessitates an experienced multidisciplinary team and adherence to standardized forensic guidelines, given that misinterpretation of VUS may lead to inaccurate genetic diagnoses and inappropriate therapeutic interventions in surviving relatives^[70].

We summarized representative SUD-related genes primarily on the basis of ClinGen gene-disease validity classifications, prioritizing genes with definitive, strong, or moderate evidence for disease causation in Figure 3. Representative pathogenic variants in arrhythmia-associated genes and their evidence scores are shown in Table 1. Representative examples of genetic variants in KCNQ1 classified according to the ACMG/AMP five-tier system are shown in Table 2.

Figure 3. Summary of genes associated with conditions contributing to SUD^[72-80]. SUD: Sudden unexplained death; BrS: Brugada syndrome; SQTS: short QT syndrome; CPVT: catecholaminergic polymorphic ventricular tachycardia; LQTS: long QT syndrome; DS: Dravet syndrome; DEE: developmental and epileptic encephalopathy; ARVC: arrhythmogenic right ventricular cardiomyopathy; ACM: arrhythmogenic cardiomyopathy; HCM: hypertrophic cardiomyopathy; DCM: dilated cardiomyopathy.

Proteomics

Proteomics is the systematic study of protein expression, structure, and function within biological systems, and can reflect the underlying pathophysiological state prior to death and therefore hold significant forensic relevance^[123]. Over the past decades, proteomic technologies have evolved from early separation-based approaches to high-throughput analytical platforms centered on mass spectrometry.

Protein Biomarkers

Protein biomarkers remain the cornerstone of biochemical investigation in SCD. Commonly used markers include cTnI and cTnT, creatine kinase-MB (CK-MB), heart-type fatty acid-binding protein (H-FABP), glycogen phosphorylase isoenzyme BB (GPBB), lactate dehydrogenase (LDH), myoglobin, ischemia-modified albumin (IMA), and inflammatory markers such as C-reactive protein (CRP)^[123]. Among these, cardiac troponins are widely regarded as the gold standard for detecting myocardial necrosis due to their high cardiac specificity and sensitivity, with demonstrated diagnostic value in both pericardial fluid and peripheral blood^[128-130]. CK-MB, although limited by lower specificity due to skeletal muscle interference, may provide additional diagnostic value when interpreted in combination with troponins^[131,132]. Early ischemic markers such as H-FABP and GPBB are particularly useful in cases with very short survival times, as they are rapidly released following myocardial injury^[133]. The interpretation of these biomarkers in postmortem settings is complicated by factors such as postmortem interval, hemolysis, and sample heterogeneity. Consequently, a multi-marker approach is generally considered more reliable than single-analyte analysis. In recent years, novel biomarkers of myocardial ischemia have been reported, but their application in postmortem forensic investigations remains unsupported by empirical data.

Galectin-3 (Gal-3) is a β-galactoside-binding lectin secreted primarily by activated macrophages and is associated with inflammation and myocardial fibrosis. Elevated Gal-3 levels have been linked to adverse remodeling and poor prognosis in chronic HF. Gal-3 may serve as an indicator of underlying fibrotic cardiac remodeling, offering insight into structural substrates that predispose to fatal arrhythmias^[134,135].

Copeptin is released in equimolar amounts with arginine vasopressin (AVP) during physiological stress. Copeptin has been proposed as a marker of acute endogenous stress response and, when combined with cardiac troponins, may improve the early detection of acute myocardial events, although its standalone diagnostic value remains limited^[136].

Cardiac myosin-binding protein C (cMyBP-C) is a structural protein of the sarcomere that plays a key role in regulating myocardial contraction. It is rapidly released into circulation following myocardial injury and has demonstrated high cardiac specificity. Recent studies suggest that cMyBP-C may be a sensitive biomarker for early myocardial damage^[60,66], potentially outperforming traditional markers in certain contexts, making it a promising candidate for forensic diagnosis of acute cardiac death.

Selenium-binding protein 1 (SELENBP1) and vinculin (VCL) have recently been investigated in cases of sudden death associated with coronary artery spasm^[137].

SELENBP1 is involved in cellular redox regulation, while VCL plays a role in cytoskeletal integrity and cell adhesion. Studies have demonstrated that SELENBP1 exhibits high sensitivity and specificity, whereas VCL provides additional discriminatory value, suggesting their potential utility as novel biomarkers in specific subtypes of SCD^[137].

Pentraxin-3 (PTX3) is an acute-phase protein produced by macrophages and vascular endothelial cells in response to inflammation, reflecting local vascular inflammation and associated with atherosclerotic plaque instability and thrombosis. Elevated PTX3 levels indicate acute coronary events^[137,138], particularly in cases where histological evidence is limited, providing additional support for the diagnosis of ischemic death.

Metabolomics

Metabolomics focuses on the comprehensive analysis of small-molecule metabolites within biological systems, typically reflecting the downstream outcomes of gene expression and protein activity. Current metabolomics studies in death and diseases are primarily based on analytical platforms such as LC-MS, gas chromatography-mass spectrometry (GC-MS), and nuclear magnetic resonance (NMR), which enable the detection and quantification of a broad range of endogenous metabolites^[139-141]. These techniques allow for the characterization of metabolic alterations associated with cardiac dysfunction, hypoxia, electrolyte imbalance, and other pathophysiological processes that may contribute to sudden death.

Hypoxia/ischemia (e.g., cerebral ischemia, myocardial infarction) induces metabolic dysfunction, alters amino acid profiles (e.g., decreased glycine, isoleucine, lysine), shifts metabolism to anaerobic glycolysis, causing lactate and pyruvate accumulation in postmortem fluids, and triggers oxidative stress and ROS^[142,143]. These metabolites serve as key evidence supporting diagnosis of ischemic causes of death and asphyxia^{[123,144-147]}. Though non-specific, these changes support forensic diagnosis of systemic stress and ischemia when interpreted alongside other metabolic markers^[148,149].

Metabolomics provides functional molecular evidence for SUD cases with negative conventional autopsies by capturing terminal metabolic fingerprints of hypoxia, oxidative stress, and systemic inflammation. Although individual metabolite changes lack specificity, integrating multi-marker panels through machine learning (ML) algorithms and combination with molecular autopsy can effectively support the differential diagnosis of ischemic causes of death, asphyxia, and endogenous stress-triggered events^[150]. Specific metabolite panels have shown promise in distinguishing SCD from other causes of death. For instance, alterations in the kynurenine pathway (e.g., increased kynurenine/tryptophan ratio) have been observed in SCD cases, reflecting inflammation and oxidative stress^[151,152]. A recent study identified a combination of pyruvate, L-threonic acid, and N-acetyl-L-cysteine as a potential signature for ACS-related sudden death, with improved discriminatory power compared to any single metabolite^[153]. ML has been increasingly integrated with metabolomics to enhance forensic classification. For example, supervised ML algorithms such as random forest and support vector machines (SVM) have been trained on postmortem metabolomics profiles to discriminate between SCD, asphyxia, and poisoning, achieving high accuracy^[154,155]. A large-scale study of 4,282 human postmortem cases demonstrated that orthogonal partial least squares discriminant analysis (OPLS-DA) models could discriminate five COD groups (acidosis, drug intoxication, hanging, ischemic heart disease, and pneumonia) with specificity-optimized prediction accuracies ranging from 73% to 99%^[156]. These approaches can identify complex, non-linear metabolic signatures that are not detectable by univariate analysis. In SUD cases, ML-assisted metabolomics may help uncover distinct metabolic fingerprints associated with channelopathies, concealed cardiomyopathies, or epileptic seizures, offering a novel avenue for postmortem diagnosis. Metabolomics facilitates a multi-omics integration framework that transforms SUD from a diagnosis of exclusion into a precision-based assessment supported by objective pathophysiological evidence, addressing critical gaps in conventional forensic evaluation of SUDs.

Transcriptomics

Transcriptomics allows the investigation of functional changes at the gene expression level. Dysregulation of genes encoding ion channel proteins may contribute to electrical instability and increase susceptibility to fatal arrhythmias^[157]. In addition, transcriptional changes associated with fibrosis, inflammation, and apoptosis suggest that subtle molecular remodeling may precede overt structural abnormalities detectable by conventional autopsy^[158].

Non-coding RNAs - particularly miRNAs - have emerged as highly promising candidates for forensic applications. Due to the tissue specificity and relative stability in postmortem tissues and body fluids, miRNAs serve as ideal biomarkers in forensic investigations. Several cardiac-enriched miRNAs, such as miR-1, miR-133, miR-208, and miR-499, have been shown to be significantly upregulated in cases of ischemic sudden death and correlate with established markers of myocardial injury and histological evidence of necrosis^[159,160]. Importantly, distinct miRNA expression profiles may enable differentiation between ischemic SUD and primary arrhythmic deaths without structural abnormalities, providing valuable support in autopsy-negative cases^[160]. High-throughput sequencing and quantitative PCR studies have demonstrated that miRNA profiles remain stable for up to 196 h postmortem, reinforcing their reliability as postmortem biomarkers^[161]. Moreover, the analysis of extracellular vesicles has shown enhanced diagnostic performance compared to circulating miRNAs alone^[18,162]. Encapsulation within vesicles protects nucleic acids from degradation and may preserve disease-specific molecular signatures, enabling improved discrimination between ischemic and non-cardiac COD, as well as between different cardiac pathologies.

Current available biomarkers of SCD, SIDS are listed in Table 3. Although no specific biomarker for SUDEP has been reported so far, several promising candidates are under active investigation. Potential biomarkers include circulating neuropeptides (e.g., substance P, neuropeptide Y) reflecting autonomic dysregulation, inflammatory markers (e.g., IL-6, TNF-α) associated with seizure-induced neuroinflammation, and miRNA signatures (e.g., miR-134, miR-146a) linked to neuronal excitability and seizure severity^[173-175]. Future multi-omics studies are expected to identify and validate SUDEP-specific biomarkers that could aid in risk stratification and forensic diagnosis.

Raman spectroscopy

Raman spectroscopy is a label-free vibrational spectroscopic method that reflects the biochemical composition of biological samples through inelastic light scattering^[176]. Surface-enhanced Raman spectroscopy (SERS) is more suitable for detecting low-abundance targets in complex biological samples for its sensitivity^[177]. In SUD, Raman-based analysis may help identify subtle biochemical changes^{[123,178,179]}, and may be used for rapid molecular profiling of tissues and body fluids and therefore complement conventional biomarkers^[123,176]. Unlike conventional assays that focus on individual analytes, Raman spectra reflect broader molecular changes and may provide a more integrated view of the perimortem biochemical state^[180]. This is consistent with recent studies of SERS-based cardiac biomarker profiling, in which multiplex platforms have enabled simultaneous assessment of several serum cardiac markers and improved molecular characterization of cardiovascular risk states^{[177,181,182]}.

Our recent studies have shown that SERS-based platforms can be used for cross-category biomarker detection and early warning of SCD^[183]. These findings indicate that Raman-derived molecular fingerprints may be useful in the combined assessment of different biomarker classes in sudden cardiac events^[183,184]. At present, however, the forensic application of Raman spectroscopy in SUD remains limited. Major constraints include matrix interference, postmortem degradation, signal reproducibility, and the lack of standardized forensic reference databases^[123,178]. Raman spectroscopy is therefore better regarded as a supplementary technique than as an independent diagnostic method in the forensic investigation of SUD^[185,186].

Fourier transform infrared spectroscopy techniques

Fourier transform infrared (FTIR) spectroscopy techniques have emerged as a powerful, label-free tool for forensic death investigation by capturing subtle biochemical alterations that are often undetectable by conventional autopsy. It has been successfully applied to identify key biomarkers in a restraint stress model, and differentiate COD such as asphyxia from sudden cardiac arrest^[187]. When combined with ML algorithms, FTIR enables objective and high-throughput classification of postmortem samples, offering a complementary approach to traditional histology^[188-190]. Furthermore, the application of attenuated total reflection (ATR)-FTIR to biofluids (serum, plasma, urine) provides a non-invasive “fluid biopsy” strategy to assess systemic metabolic disturbances^[188,191], making it particularly valuable for cases where tissue samples are limited or ambiguous, such as in SUD. These integrated spectroscopic and computational strategies hold promise for transforming forensic pathology by providing rapid, objective, and molecularly grounded evidence for COD determination.

PMCT

Postmortem computed tomography (PMCT) has become one of the most widely used imaging modalities in forensic practice^[192]. Because SUD is the absence of a definitive COD after conventional autopsy and histological examination, the primary value of PMCT lies in refining the diagnostic process through exclusion, guidance, and supportive findings. The key contributions of PMCT in SUD investigations are the systematic exclusion of alternative COD that may be subtle, atypical, or easily overlooked during conventional autopsy. Postmortem gas formation, livor mortis, tissue density changes, vascular integrity loss, and fluid redistribution can all alter CT imaging characteristics over time, potentially leading to misinterpretation^[192,193]. In SUD cases, especially those involving suspected infection or metabolic disturbances such as diabetic ketoacidosis, specific gas distribution patterns may provide important clues^[193,194]. PMCT also plays an important role in excluding external causes that may mimic SUD. For example, in drowning cases, PMCT may demonstrate characteristic findings such as fluid accumulation in the airways and paranasal sinuses, ground-glass opacities in the lungs, pulmonary hyperinflation, and retention of foam-like material within the airways^[195]. In cases where arrhythmogenic cardiomyopathies or other subtle myocardial abnormalities are suspected, PMCT may reveal indirect or localized changes in myocardial structure, such as areas of altered density that could correspond to fibrofatty replacement^[196,197]. Although these findings are often insufficient for definitive diagnosis, they can provide valuable guidance for focused histological sampling and molecular investigations, including genetic testing. This targeted approach is particularly important in SUD, where diffuse or non-specific sampling may fail to detect localized pathological substrates. In this way, PMCT contributes to a more structured and hypothesis-driven investigation process.

Despite these advantages, PMCT has limited sensitivity for soft tissue pathology and early-stage disease processes, and it cannot replace conventional autopsy in determining the COD in SUD cases. The value of PMCT in SUD investigations is not in directly diagnosing the COD, but in excluding alternative explanations, guiding targeted examinations, and providing supportive evidence. This integrative role makes PMCT a crucial component of modern forensic workflows, particularly in cases where the COD remains elusive after routine examination.

PMCTA

Postmortem computed tomography angiography (PMCTA) represents an important extension of PMCT, enabling detailed visualization of the vascular system through the administration of contrast agents. The primary contribution of PMCTA in SUD investigations is the detection and exclusion of vascular COD that may not be apparent on standard autopsy^[198,199]. PMCTA allows for enhanced visualization of the coronary arteries and major vessels, facilitating the identification of stenosis, occlusion, or abnormal contrast distribution patterns^{[198,200-202]}. Even when these findings are not definitive, they can help reclassify cases initially considered as SUD into more specific cardiovascular categories.

A limitation of PMCTA lies in the use of contrast agents, which can introduce artifacts such as extravasation or artificial vessel distension and may complicate interpretation. Additionally, postmortem changes in vascular tone and permeability can affect contrast distribution, potentially leading to false-positive or false-negative findings^[198]. For this reason, the timing of PMCTA and the choice of contrast medium are critical factors in ensuring reliable results. PMCTA provides a valuable adjunct in the investigation of SUD by enhancing vascular assessment, facilitating the exclusion of occult vascular causes, and supporting targeted diagnostic strategies.

PMMRI

Postmortem magnetic resonance imaging (PMMRI) represents a critical advancement in virtual autopsy, particularly in the investigation of SUD, offering superior soft tissue contrast and provides unique insights into subtle pathological changes that may underlie functionally mediated deaths^[203]. The most significant contribution of PMMRI in SUD lies in its ability to detect subtle myocardial abnormalities that are not readily identifiable through gross examination or even routine histology. In conditions such as ACM, early-stage myocardial changes may be unevenly distributed and easily missed during standard sampling^[191,204]. PMMRI enables a more comprehensive, whole-heart assessment, allowing for the identification of regional signal alterations that may correspond to underlying pathological substrates. These findings can then guide targeted histological sampling and improve the diagnostic yield of subsequent analyses.

PMMRI provides substantial advantages in the assessment of the central nervous system. In cases involving brain softening, edema, or subtle hemorrhagic changes, PMMRI has demonstrated superior sensitivity compared to both PMCT and conventional autopsy^[194,201], allowing for improved visualization of gray-white matter differentiation, detection of microhemorrhages, and characterization of liquefactive changes that may not be appreciable during dissection. These capabilities are particularly relevant in SUD cases where neurological causes, such as seizure-related death or central autonomic dysfunction, are suspected but remain difficult to confirm.

In addition, advanced MRI techniques, such as magnetic resonance spectroscopy (MRS), have been explored for postmortem biochemical analysis^[205-208]. These approaches may offer insights into metabolic conditions or substance exposure prior to death. However, their current forensic applicability remains limited due to variability and insufficient accuracy, and they should be regarded as experimental rather than routine tools in SUD investigation.

Machine learning

ML has emerged as a particularly adaptable tool for handling complex and heterogeneous data^[209]. Given its characteristic of allowing patterns to be learned directly from empirical observations, ML has been introduced in the COD identification in forensic practice, where the diagnostic process is often challenged by the absence of definitive structural findings at conventional autopsy^[210]. In such cases, ML offers a framework capable of integrating diverse forms of evidence, thereby providing a more comprehensive basis for forensic interpretation.

From a methodological perspective, ML approaches can be broadly divided into supervised and unsupervised learning. Supervised learning involves training models on labeled datasets to perform tasks such as classification or regression, whereas unsupervised learning focuses on identifying intrinsic structures within unlabeled data, such as clustering or dimensionality reduction^[211]. In SUD cases, the central issue is the gradual reduction of uncertainty through the integration of weak or indirect signals^[210]. ML can detect patterns that are not immediately apparent through conventional statistical or visual analysis, thereby transforming fragmented observations into more coherent biological or forensic narratives.

One of the most mature areas for ML application in death and diseases is omics-based analysis^[212]. Postmortem biological samples, such as cardiac blood, peripheral blood, or myocardial tissue, can yield high-dimensional datasets through multi-omics researches. In SUDs, metabolic disturbances - such as those associated with early ischemia, ion channel dysfunction, or systemic metabolic imbalance - tend to manifest as subtle, coordinated changes across multiple metabolites^[213], where multivariate patterns can be captured by ML algorithms. For instance, supervised models such as random forest or SVM can be trained to distinguish SUD cases from controls by identifying combinations of metabolites that collectively contribute to classification^[214]. In practical terms, ML-assisted metabolomics analysis can help forensic investigators move beyond descriptive findings toward more nuanced interpretations, such as the presence of metabolic signatures consistent with arrhythmogenic mechanisms or systemic physiological stress^[123,215]. Although these findings may not be definitive on their own, they contribute to a cumulative evidentiary framework that supports COD assessment in otherwise unexplained cases.

A second major domain where ML demonstrates significant utility is molecular autopsy^[216], particularly in the interpretation of genetic data. Advances in NGS have introduced a new challenge of identifying numerous VUS^[217,218], many of which cannot be easily classified as pathogenic or benign using conventional criteria. More importantly, ML enables a shift from single-case interpretation to cohort-level analysis. By analyzing data across multiple SUD cases, unsupervised learning methods such as clustering can identify patterns of variant co-occurrence or shared genetic profiles. This is particularly valuable in conditions like channelopathies, where diagnosis relies heavily on molecular evidence. In addition, ML can facilitate the integration of genetic data with phenotypic and circumstantial information, such as age, sex, family history, and scene findings. This multi-layered approach reflects the reality of forensic practice, where conclusions are rarely based on a single line of evidence^[217].

The application of ML to virtual autopsy represents a more recent but rapidly evolving area^[219,220]. Postmortem imaging techniques, particularly PMCT and PMMRI, generate large volumes of data that are often underutilized due to the complexity of interpretation and the presence of postmortem artifacts^[193,221]. In SUD cases, imaging findings are frequently subtle and nonspecific, which limits their standalone diagnostic value. ML can enhance the utility of these data by providing tools for pattern recognition and quantitative analysis^[222]. For example, supervised learning models can be trained to identify imaging features associated with specific pathological conditions^[205]. Moreover, ML-based approaches can provide consistent criteria for identifying and quantifying features, thereby avoiding the variability between observers^[222].

A critical barrier to the widespread application of ML in forensic SUD investigation is the lack of large-scale, multi-center, standardized datasets. Unlike clinical medicine, where initiatives such as TCGA and UK Biobank have enabled data sharing across institutions, forensic genetics currently lacks comparable resources. Recently, a multi-center dataset comprising over 16 million high-resolution image patches, 2,228 vision-language pairs, and 471 distinct diagnostic outcomes from three premier forensic cohorts has been successfully compiled for AI model development^[223]. This example underscores the potential of shared forensic data resources. Therefore, the establishment of multi-center shared datasets with standardized protocols for sample collection, sequencing, and data annotation is urgently needed to train robust ML models and enable cross-population validation.

However, data sharing in forensic genetics faces substantial privacy and legal barriers. Studies have shown that as few as 10 common SNPs are sufficient to identify individuals from a reference database of ~5,000 haplotypes, and 20-30 SNPs can identify first-order relatives, demonstrating that genetic data are inherently re-identifiable even when anonymized^[224]. This raises serious questions about whether "de-identified" forensic DNA profiles can truly guarantee anonymity^[225]. The legal basis for accessing genetic databases for research or investigative purposes remains ambiguous. Cross-border data sharing is further complicated by jurisdictional conflicts between divergent national privacy laws and a lack of harmonized international legal frameworks^[226]. Jurisdictional conflicts arise from divergent national privacy laws, and concerns over function creep, where data collected for humanitarian purposes (e.g., missing persons identification) may be repurposed for law enforcement, have been raised by multiple stakeholders^[226]. Furthermore, secondary use of biobank samples for forensic purposes without explicit consent remains under-regulated^[225]. These challenges underscore the need for harmonized legal frameworks, transparent governance, and stakeholder engagement to enable responsible data sharing without compromising individual privacy rights.

The integration of ML into SUD investigation reflects a broader shift in forensic medicine from descriptive to data-driven analysis. ML serves as a complementary tool that enhances the interpretation of complex datasets. By bridging the gap between omics, molecular autopsy, and virtual autopsy, ML contributes to a more holistic understanding of SUD, ultimately supporting more informed and transparent forensic conclusions.

Deep learning

Deep learning (DL), as an advanced branch of ML, has introduced a fundamentally different approach to data analysis by enabling models to automatically learn hierarchical representations from raw input data^[216]. Unlike traditional ML methods that rely on manually selected features, DL architectures - particularly convolutional neural networks (CNNs), recurrent neural networks (RNNs), and related variants - are capable of extracting complex, non-linear patterns directly from high-dimensional datasets^[216].

DL improves data from molecular autopsy by modeling complex, nonlinear relationships across large-scale genomic datasets, functional annotations, and evolutionary features, thereby enhancing the accuracy and depth of pathogenicity prediction for both coding and non-coding variants. For instance, deleterious annotation of genetic variants using neural networks (DANN) has been developed employing a DL architecture trained on functional annotations similar to those used in combined annotation-dependent depletion (CADD), enabling improved discrimination between benign and pathogenic variants^[227]. By capturing nonlinear interactions among diverse genomic features, such models provide probabilistic assessments of variant pathogenicity even in the absence of direct experimental evidence. In SUD cases, this is particularly valuable for prioritizing rare or novel variants that cannot be confidently classified using conventional guidelines such as those from the ACMG. Similarly, PrimateAI incorporates evolutionary information derived from primate genomes to evaluate the functional impact of missense variants^[228], leveraging cross-species conservation to identify potentially deleterious substitutions. DeepSEA and Basenji utilize CNNs to predict chromatin accessibility, transcription factor binding, and gene expression directly from DNA sequences^[229,230]. Additionally, PromoterAI further refines this approach by specifically predicting the impact of promoter variants^[231], enhancing the resolution of regulatory variant interpretation. SpliceAI leverages deep CNNs to capture long-range dependencies within pre-mRNA sequences, enabling accurate prediction of splice site disruptions caused by genetic variants. This is particularly relevant in forensic cases, as certain variants previously classified as synonymous or benign may in fact exert pathogenic effects by altering RNA splicing^[232,233]. In practice, these predictions can be integrated with RNA sequencing or in vitro functional assays to establish a more robust evidentiary basis.

DL has also shown immediate impact in virtual autopsy, particularly in the analysis of PMCT and PMMRI data. DL models can be trained to detect patterns associated with SUDs such as ACM, including diffuse fatty infiltration or subtle myocardial remodeling, which may not be apparent on routine visual inspection^[222]. Another practical advantage of DL in virtual autopsy is its ability to enable quantitative and standardized analysis^[219]. In SUD cases, where findings may be borderline or non-specific, such quantitative approaches can add an additional layer of evidence that supports or refines forensic conclusions.

Despite these promising applications, it is important to recognize that DL in SUD investigation remains at an early stage, and several challenges must be addressed before widespread adoption. One of the main limitations is the availability of high-quality, well-annotated datasets, which are essential for training reliable models. In forensic medicine, data are often limited in size and subject to variability in acquisition and documentation. In addition, the “black box” nature of many DL models raises concerns about interpretability, particularly in legal contexts where transparency and reproducibility are critical. For this reason, there is increasing interest in developing explainable DL approaches that provide insight into how predictions are made, thereby enhancing trust and usability in forensic applications. By enabling the analysis of complex, high-dimensional data in a more integrated and automated manner, DL supports a shift toward more data-driven and system-level interpretations of SUD. When combined with traditional forensic methods, it has the potential to improve diagnostic yield, reduce uncertainty, and provide a more comprehensive understanding of cases that would otherwise remain unexplained [Figure 4].

Figure 4. Integrated analytical framework for the application of AI in forensic investigation of SUDs. This schematic illustrates the multi-modal strategy combining multi-omics and spectroscopic approaches, virtual autopsy imaging techniques, and AI (ML and DL) for precise COD determination, pathophysiological mechanism interpretation, and evidence-based diagnosis in SUD cases (including SCD, SIDS, and SUDEP). Created in BioRender. qi, M. (2026) https://BioRender.com/oepjqkf. SUD: Sudden unexplained death; NGS: next-generation sequencing; SERS: surface-enhanced Raman spectroscopy; FTIR: Fourier transform infrared; DANN: deleterious annotation of genetic variants using neural networks; CNN: convolutional neural network; SVM: support vector machine; AI: artificial intelligence.