Determining how long a person has been dead is one of the most stubborn problems in forensic science, and a new comprehensive review argues that the answer may lie in a surprisingly stable corner of the human body: the vitreous humor, the clear gel that fills the eyeball. In a paper published in the International Journal of Legal Medicine, researchers from Kunming Medical University synthesize decades of research on postmortem interval estimation using vitreous humor analysis, and chart a path toward a future in which machine learning models fuse multiple biochemical signals into precise, courtroom-ready estimates of time since death. The work arrives at a moment when forensic investigators worldwide are increasingly frustrated by the limits of traditional methods, which can be thrown off by weather, clothing, body size and a host of other confounders.
The conventional toolbox for estimating the postmortem interval, or PMI, has long relied on physical signs of decomposition: body temperature cooling curves, the onset and resolution of rigor mortis, lividity patterns, and in later stages, insect colonization and microbial succession. Each of these approaches has well-documented weaknesses. Ambient temperature swings, morgue storage, wrappings and encasement, and individual variation in physiology can all distort the timelines inferred from these markers. Researchers have even shown that rigor mortis can persist far longer than classic models assume when bodies are kept cold, and that maggot development continues during refrigerated storage, complicating entomological estimates. The result, as the review’s authors note, is that conventional methods are often compromised by environmental factors and variable preservation conditions, restricting their accuracy and practical utility.
Vitreous humor offers a way around many of these problems, and the reason is anatomical. The fluid is sequestered behind the blood-retinal barrier, a selective gatekeeper that tightly controls which substances pass between the bloodstream and the eye. Because of this isolation, the vitreous humor exhibits relative metabolic stability after death and resists contamination and bacterial invasion far better than blood, cerebrospinal fluid or tissue samples. It is also less susceptible to the postmortem redistribution of drugs and metabolites that plagues toxicological interpretation. These properties make it, in the words of the review, an optimal biological matrix for PMI estimation, one that can extend the effective time window during which reasonably precise estimates are possible.
The scientific pedigree of vitreous humor analysis stretches back more than a century, to structural studies of the gel in the 1860s, but its forensic career truly began in the 1960s. In 1963, investigators reported that potassium concentrations in the vitreous humor rise steadily after death, a finding quickly confirmed by parallel studies showing that the correlation could serve as a clock for the postmortem interval. The mechanism is elegantly grim: after circulation stops, cells lose their ability to maintain ion gradients, and potassium leaks from intracellular stores into the surrounding fluid at a roughly predictable rate. Later work added hypoxanthine, a breakdown product of ATP metabolism that accumulates as oxygen deprivation sets in, and showed that combining hypoxanthine with potassium and ambient temperature improves estimates further.
Beyond potassium and hypoxanthine, the review catalogues a rich cast of biochemical characters whose postmortem trajectories correlate with time since death. Sodium and chloride concentrations shift in patterns that have proven useful both for PMI estimation and for distinguishing saltwater drowning from immersion deaths unrelated to drowning. Calcium and magnesium rise as cellular membranes fail. Glucose, urea nitrogen, creatinine and uric acid provide complementary information, though glucose is complicated by the fact that it can be elevated in conditions such as ethylene glycol poisoning. Ammonium accumulates as proteins degrade, and has even been measured at crime scenes using microfluidic paper-based devices designed to bring thanatochemistry out of the autopsy suite and into the field. Free amino acids, lactate, and protein fragments analyzed by high-performance liquid chromatography and mass spectrometry round out an expanding analytical repertoire.
Historically, the mathematical treatment of these signals relied on linear regression equations, most famously the potassium-based formulas refined across generations of studies. These equations correct for ambient temperature, age and other covariates, and software tools have been built to make them accessible to practicing pathologists. Yet the underlying biology is not strictly linear. Electrolyte diffusion, enzymatic degradation, temperature-dependent reaction kinetics and individual variation all interact in complicated ways, and single-marker equations carry confidence intervals that can span hours or days. Flexible regression models and chemometric approaches have pushed accuracy further, but the review argues that the real leap forward comes from artificial intelligence.
Machine learning entered the field in earnest in 2002, when researchers paired capillary zone electrophoresis measurements of vitreous electrolytes with artificial neural networks to estimate the PMI, demonstrating that nonlinear models could capture dependencies that simple equations missed. Since then, the approach has gathered momentum. Neural networks and other algorithms have been applied to sodium and potassium data, to metabolomic profiles of ocular fluids, and to microbiome succession data from animal models. Recent studies combining the human metabolome with machine learning have reported substantially improved PMI predictions, and systematic reviews now document machine learning and metabolomics applications across multiple tissue types. The review emphasizes that these models do more than enhance predictive accuracy; they enable robust nonlinear time series analysis of the complex, interacting biochemical changes that unfold after death.
The authors’ forward-looking vision is a multimodal fusion framework that integrates multiple streams of information: vitreous electrolytes, hypoxanthine and lactate kinetics, amino acid and peptide degradation patterns, imaging data such as the postmortem changes visible in the eye on computed tomography, and contextual variables like temperature and body weight. By training machine learning models on such heterogeneous datasets, the framework would hedge against the weaknesses of any single marker while exploiting the complementary strengths of many. The review also points to enabling technologies that could make this vision practical at scale, including tandem mass spectrometry, capillary electrophoresis, electrochemical biosensors for hypoxanthine, and rapid field-deployable microfluidic devices. Such a framework, the authors argue, is poised to become an indispensable tool for PMI estimation in forensic investigations and medicolegal proceedings.
Challenges remain before the vision becomes routine practice. Machine learning models are only as good as the data they are trained on, and the forensic literature contains decades of measurements gathered with different instruments, protocols and populations. Standardization of sampling and analytical methods will be essential, as will validation across diverse climatic conditions, causes of death and antemortem physiological states, since factors such as pre-death hypoxia can independently elevate hypoxanthine levels. Interpretability is another concern; courts and defense attorneys will demand to know why a model produced a particular estimate. Still, the review makes a compelling case that the convergence of a uniquely stable biological matrix, a maturing analytical toolkit and powerful computational methods is transforming one of forensic science’s oldest quests. What began with a simple potassium measurement in the 1960s is evolving into a data-rich, algorithm-driven science of the dead, one in which the fluid of the eye may soon testify, with unprecedented precision, about the moment life ended.
The choice of vitreous humor over alternative biological matrices reflects a broader pattern in postmortem biochemistry, where researchers have explored pericardial fluid, cerebrospinal fluid, and even synovial fluid as potential sources of time-dependent markers. Each fluid has its own advantages and drawbacks. Cerebrospinal fluid, for instance, contains promising markers such as glial fibrillary acidic protein, neuron-specific enolase and S100B, but its biochemical composition can vary depending on the sampling site along the spinal column, a source of variability that complicates comparisons across studies. Pericardial fluid has proven valuable for metabolomic investigations and for detecting markers of antemortem drug exposure, yet it is more exposed to contamination and postmortem diffusion from adjacent tissues than the sequestered ocular environment.
Practical considerations also favor the eye in casework. Vitreous humor can be collected with a simple needle aspiration through the sclera, typically from both eyes, and the small sample volumes required are compatible with standard clinical biochemistry analyzers as well as more sophisticated platforms. The fluid’s clarity and low protein content reduce matrix effects in spectrophotometric and chromatographic assays, which is one reason potassium-based methods achieved early reproducibility. At the same time, forensic researchers have cautioned that sampling technique, the interval between collection and analysis, and storage temperature can all influence measured concentrations, underscoring the need for harmonized protocols.
Imaging-based approaches add a complementary, non-destructive dimension. Studies of postmortem computed tomography have documented progressive changes in the eye, and measurements of the radiodensity of vitreous humor and cerebrospinal fluid have been proposed as indicators of the time since death. Because such imaging can be performed before autopsy without altering the specimen, radiological markers could eventually be integrated alongside biochemical ones in the multimodal frameworks the review envisions.
The pursuit of a single universal formula for the postmortem interval has long been viewed with skepticism among forensic scientists, and this skepticism shapes the review’s argument for data-driven fusion. Rather than seeking one equation that fits all circumstances, the field appears to be moving toward context-aware models that weigh temperature history, individual physiology and multiple analytes simultaneously. If validated rigorously, such models would not replace the forensic pathologist’s judgment but would supply a quantified, uncertainty-aware estimate that can be communicated transparently in medicolegal settings, marking a meaningful shift in how time since death is established in practice.
Subject of Research: Postmortem interval estimation using vitreous humor biochemistry and machine learning
Article Title: From biochemical markers to machine learning: a review of postmortem interval estimation via vitreous humor analysis
Article References: Rao, M., Wang, C., Tian, Y., Tao, H., Liu, L., & Zeng, X. (2026). From biochemical markers to machine learning: a review of postmortem interval estimation via vitreous humor analysis. International Journal of Legal Medicine. https://doi.org/10.1007/s00414-026-03979-8
Image Credits: AI Generated
DOI: 10.1007/s00414-026-03979-8
Keywords: vitreous humor, postmortem interval, forensic science, machine learning, potassium, hypoxanthine, blood-retinal barrier, thanatochemistry, forensic toxicology, artificial neural networks, multimodal fusion, death investigation
Cite Scienmag News
APA
MLA
Chicago
Teresa Odom. (September 12, 2026). Eye Fluid Chemistry and Machine Learning May Pinpoint Time of Death. Scienmag. https://scienmag.com/eye-fluid-chemistry-and-machine-learning-may-pinpoint-time-of-death/
Teresa Odom. “Eye Fluid Chemistry and Machine Learning May Pinpoint Time of Death.” Scienmag, 12 September 2026, https://scienmag.com/eye-fluid-chemistry-and-machine-learning-may-pinpoint-time-of-death/. Accessed 12 September 2026.
Teresa Odom. “Eye Fluid Chemistry and Machine Learning May Pinpoint Time of Death.” Scienmag. September 12, 2026. https://scienmag.com/eye-fluid-chemistry-and-machine-learning-may-pinpoint-time-of-death/
Copy citation
Download RIS
Tags: advances in legal medicineartificial neural networksbiochemical signals in eye fluidblood-retinal barrierchallenges of traditional PMI methodscourtroom forensic techniquesdeath investigationenvironmental effects on postmortem intervalforensic biomarkersforensic scienceforensic toxicologyhypoxanthineMachine learningmachine learning in forensic medicinemicrobial succession in decompositionmultimodal fusionpostmortem intervalpostmortem interval estimationpotassiumthanatochemistrytime of death determinationvitreous humorvitreous humor analysis
