Mapping the hidden journey of microplastics: Multi-organ deposition patterns and organ-specific health risks revealed by AI-driven analysis

Abstract

Microplastics (MPs), pervasive environmental pollutants, have infiltrated human tissues, raising global health concerns. This study investigated the distribution and characteristics of MPs across seven major human organs (lungs, heart, liver, spleen, brain, kidneys, and small intestine) using Raman imaging and machine learning. Tissue samples from eight donors were analyzed for MP presence and characteristics. A deep learning-enhanced U-Net model segmented MPs in Raman images, while a random forest classifier was employed to identify organ-specific MP attribution using 120 imaging features. Animal models supported the systemic distribution of MPs. MPs were ubiquitous across all organs examined. The highest MP abundance was observed in the liver (65.28 ± 23.94 particles/g), small intestine (61.06 ± 25.25 particles/g), and kidneys (58.63 ± 16.50 particles/g). Organ-specific variations in MP characteristics were identified: larger particles dominated the lungs (56.80 ± 57.70 μm), while smaller particles (<10 μm) prevailed in the liver and spleen. Distinct polymer compositions and shape profiles were observed for each organ. The random forest classifier achieved 72.73% accuracy in organ-specific MP attribution. MP abundance was linked to organ vascularity. The findings highlight organ-specific risks of MPs and provide a framework for assessing health impacts, thus guiding targeted interventions to mitigate exposure.

Graphical abstract

Public summary

• •Microplastics (MPs) were detected in all seven human organs, highest in liver, small intestine, and kidneys.
• •Organ-specific MP traits; lungs had larger particles, liver/spleen dominated by <10 μm MPs.
• •Random forest classifier achieved 72.73% accuracy in organ-specific MP attribution.

Introduction

Microplastics (MPs; <5 mm in size), as global environmental pollutants, have emerged as one of the most pressing public health concerns of the 21^st century because of their ubiquity and potential health risks.¹ With exponential growth in plastic production and inadequate waste management systems,² MPs have permeated Earth’s ecosystems and entered the human body via food chains,³ inhalation,⁴ and dermal contact.⁵ Recent studies have detected MPs in human lung tissue,⁶ blood,⁷ liver,⁸ placenta,⁹ bone,¹⁰ and even the brain,¹¹ demonstrating their alarming ability to breach biological barriers. Long-term exposure to MPs has been linked to oxidative stress,^12,13 inflammatory responses,^14,15 cellular dysfunction,^16,17 and synergistic toxicity^18,19 due to their characteristic ability to carry other pollutants. These findings have not only deepened scientific inquiry into the mechanisms of MPs toxicity but have also sounded an urgent alarm for human health crises.

Existing studies on MPs in humans have often focused on single organs, leaving the distribution patterns, organ-specific characteristics, and health risk correlations across major organs poorly understood. Key challenges include limited sample availability (relying on rare multi-organ donations) and the physicochemical diversity of MPs, such as size (1 μm–5 mm), shape (fibers, fragments, and spheres), polymers (polyethylene [PE], polypropylene [PP], and polystyrene [PS]). Conventional statistical methods have difficulty identifying patterns in high-dimensional data.

This study analyzed tissue samples from eight donors, employing Raman imaging and artificial intelligence (AI) to elucidate MP distribution and characteristics across seven organs. Utilizing a deep learning-enhanced U-Net model, MPs were segmented in Raman images, and the organ-specific MP attribution was identified using a random forest classifier and 120 imaging features. Animal experiments were performed to validate systemic MP distribution patterns. By integrating deep learning into MP image analysis, this work overcomes traditional statistical limitations, establishes the first organ-specific MP biodistribution prediction model, and bridges environmental exposure with internal deposition. This framework addresses critical gaps in MP biodistribution research and provides a methodological foundation for toxicological studies and public health policymaking.

Materials and methods

Sample collection

This study was conducted following the Declaration of Helsinki and approved by the Ethics Committee of the Fourth Military Medical University Xijing Hospital (KY20242164-C-1). All human tissue samples were obtained from registered donors at the Xi’an Red Cross Body Donation Center (Fourth Military Medical University Branch) who consented to the use of their bodies for medical education and scientific research. Tissue samples from seven organs (lungs, heart, liver, spleen, brain, kidneys, and small intestine) were collected from eight donors. Each organ specimen weighed approximately 10–20 g with a total of 56 samples. The samples were cut into small pieces for subsequent processing.

Sample processing and Raman microscopy

The sample processing was performed following a previous protocol.¹⁰ MP analysis was conducted using a Thermo Scientific DXR3xi Raman imaging system with the following parameters: 785 nm laser, 15 mW power, 0.5000 s exposure time (2 Hz), and 50 scans per point. Spectra (300–3,500 cm⁻¹) were compared against a custom spectral library (30+ polymers, covering >99% of plastic types) and the Raman Sample Library to identify polymer compositions. Particle size, shape, and color were analyzed using Nano Measurer software (v.1.2). Only those with a matching degree of over 75% were deemed valid, which offers a certain scientific basis for MP identification.¹⁰

Animal experiment design

Based on the types of MPs detected in human tissues and the widespread use of PS in animal studies, green fluorescent dye-labeled PS MP (G-PS-MP; 5 μm size, 488 nm excitation, 518 nm emission, 1% [w/v], 10 mL) was purchased from the Tianjin Basele Chromatography Technology Development Center.

Male Sprague-Dawley rats (4 weeks old) were obtained from the Fourth Military Medical University Laboratory Animal Center. The rats were housed under controlled conditions (23°C ± 1°C, 55% ± 5% humidity, 12 h light/dark cycle) with free access to food and water. After a 7-day acclimatization period, the rats were randomly divided into three groups (n = 3 per group) and administered daily intraperitoneal injections for 1 month: saline (control), 0.13 mg/kg MPs (low dose), or 1.3 mg/kg MPs (high dose). Doses were determined based on human exposure data and equivalent dose calculations. All protocols were approved by the Fourth Military Medical University Institutional Animal Care and Use Committee.

Animal specimen collection and processing

After 1 month, the rats were euthanized, and tissue samples (lungs, heart, liver, spleen, brain, kidneys, and small intestine) were collected, fixed, dehydrated, and sectioned. Fluorescence imaging was performed using an Olympus BX53 stereomicroscope (excitation wavelength: 594 nm) to detect G-PS-MPs. Images were captured with a microscope-mounted camera and analyzed using Zen imaging software (v.6.0).

Feature analysis of MPs based on Raman imaging

To investigate the distribution patterns of MPs in the human body, this study established a fully automated feature analysis pipeline for MP imaging using AI-driven methods on Raman images. The workflow comprised four core steps: target segmentation, feature extraction, feature analysis, and model construction and validation. First, standardized and data-augmented Raman images were segmented using a U-Net network enhanced with self-attention mechanisms to improve the detection accuracy of micro-scale targets. The segmentation network integrated an encoder, decoder, and skip connections to extract multi-scale features and progressively restore resolution. Boundary segmentation was optimized through a weighted combination of the Dice coefficient (L_Dice) and cross-entropy loss (L_CE), as detailed in Equations 1, 2, and 3:

$$L_{Dice}=1-\frac{2\sum pg+ϵ}{\sum p+\sum g+ϵ}$$ (Equation 1)

$$L_{CE}=-\sum \left[g\log \left(p\right)+\left(1-g\right)\log \left(1-p\right)\right]$$ (Equation 2)

$$L=\alpha L_{CE}+\beta L_{Dice}$$ (Equation 3)

Here, p represents the predicted class probability, g denotes the one-hot encoded ground truth labels, and α and β are weighting coefficients that adjust the relative importance of the two loss functions. Multidimensional features were extracted based on the segmentation results. The features included geometric shape, gray-level histogram, textural, and color features. Additionally, local binary patterns, Gabor filter features, frequency-domain features, and PyRadiomics features were incorporated. All 120-dimensional features were normalized and standardized using Z score scaling (zero mean, unit variance) to systematically analyze the feature distribution of MPs across the organs.

MP organ attribution prediction

Based on the high-dimensional feature data extracted from MP Raman images, the Isomap dimensionality reduction method was employed for nonlinear feature reduction. Isomap preserves the global geometric structure of high-dimensional data by constructing a k-nearest neighbor graph and shortest path distance matrix, achieving low-dimensional embedding. Specifically, this method first constructs a neighborhood graph of feature points and calculates the shortest path distances between data points using Floyd-Warshall or Dijkstra algorithms. Subsequently, multidimensional scaling is performed on the generated distance matrix to obtain low-dimensional embedded feature representations. The implementation is shown in Equations 4, 5, and 6:

$$\mathrm{X}=\left\{x_1,x_2,...,x_n\right\},x_i\in R^{120}$$ (Equation 4)

$$d_{ij}={\Vert x_i-x_j\Vert }_2,D={\left(D_{ij}\right)}_{n\times n}$$ (Equation 5)

$$\mathrm{Y}=\left[\sqrt{{\gamma }_1}{u}_1,\sqrt{{\gamma }_2}{u}_2,...,\sqrt{{\gamma }_d}{u}_d\right]\text{,}$$ (Equation 6)

where X represents high-dimensional feature data, $n$ is the sample size, $d_{ij}$ denotes Euclidean distance, $D$ is the geodesic distance matrix, $\mathrm{Y}$ represents the dimension-reduced data, $\Lambda =\text{diag}({\gamma }_1$, ${\gamma }_2,...,{\gamma }_n)$ is the eigenvalue matrix, and $U=[u_1$, $u_2,...,u_n$] is the corresponding eigenvector matrix. The reduced dimensional data were divided into a 70% training set and 30% test set. A random forest classification model was adopted to optimize the parameters using the training set, with the final predictions on the test set determined by majority voting among multiple decision trees, as shown in Equation 7:

$$\hat{y}={\mathit{argmax}}_k\sum _{t=1}^T\delta \left(h_t\left(x\right)=k\right)\text{,}$$ (Equation 7)

where $k$ is the class index, $\delta (·)$ is the indicator function, and $T$ subsets are used to train individual decision trees $h_t\left(x\right)$. A comparative analysis with other dimensionality reduction methods (principal-component analysis, logistic regression, and t-distributed stochastic neighbor embedding [t-SNE]) and classification approaches (support vector machine and k-nearest neighbor [KNN]) was conducted to select the optimal model for classifying MP attribution across the seven distinct organs. Finally, the system was used to analyze environmental MP Raman images and evaluate the distribution of MPs in seven specific organs.

Patient and public involvement

Our study did not include patients as study participants. No patients were involved in setting the research question or the outcome measures, nor were they involved in the design and implementation of the study.

Quality assurance and quality control

Cotton lab coats and polymer-free nitrile gloves were used throughout the analysis to eliminate potential contamination of the plastic items employed in the experiment. All sample pretreatment was performed by the same lab technician. In actual experimental research, all pretreatment procedures were carried out under a super clean bench. All glassware and instruments underwent a triple wash with ultra-pure water before being covered in aluminum foil. Prior to usage, all solvents were passed through a stainless steel filter membrane (<0.8 μm). To prevent potential air contamination, samples were counted and stored in an enclosed room. Furthermore, procedural blanks were conducted throughout the entire process to minimize the impact of background values on the measurement results.

Results

MP presence in major human organs

The present study included 8 donors, all of whom were male. Their ages at the time of death were 75, 77, 67, 72, 85, 80, 76, and 68 years, with an average age of 75.0 years. The donation period spanned from 2019 to 2022, and the storage duration ranged from 2 to 5 years. All donors resided in the central region of Shaanxi Province during their lifetime. These 8 individuals died of natural causes without a history of infectious diseases or trauma. The morphological integrity of the major organ tissues was preserved, and anti-corrosive preservation measures were implemented shortly after death.

MPs were detected in all seven major organs sampled from the human body. In total, 23 MP particles were detected in 8 brain samples, 12 MP particles were detected in 8 spleen samples, and 24 MP particles were detected in 8 small intestine samples. One sample each from kidney, liver, heart, and lung did not reveal any MPs, whereas the remaining 7 samples revealed 19, 11, 13, and 9 MP particles, respectively. Representative Raman images are shown in Figure 1. This finding underscores the ubiquitous presence of MPs in the human body. The specific distribution characteristics of MPs are as follows.

Figure 1.

Representative microphotographs and Raman spectra of the MPs

As shown in Figure 2A and Table 1, significant differences in MP abundance were observed across different human organ tissue samples. The liver, small intestine, and kidneys exhibited higher MP abundance, with mean values of 65.28 ± 23.94, 61.06 ± 25.25, and 58.63 ± 16.50 particles/g, respectively. These were followed by the brain (48.54 ± 11.20 particles/g), heart (40.34 ± 17.16 particles/g), and lungs (30.01 ± 0.49 particles/g). The spleen showed the lowest MP abundance of 12.75 ± 7.09 particles/g.

Figure 2.

Characterization of MPs in the human body

(A) Abundance of MPs in seven organs.

(B) Size of MPs in seven organs.

(C) Percentage of MPs of different particle sizes in seven organs.

(D) Percentage of each type of MPs in seven organs.

(E) Percentage of each shape of MPs in seven organs.

Table 1.

Morphology and chemical characterization of the MPs identified in human organs

Organs	Size (μm)			Abundance (particles/g)	Shape	Polymer composition
	Minimum	Maximum	Average
Heart	8.61	154.81	43.10 ± 41.29	40.34 ± 17.16	fragment, fiber, sphere	PP, PBAT, PC, PET, PPO, PP fiber, PS
Liver	2.54	8.66	5.13 ± 2.19	65.28 ± 23.94	fragment, pellet	PP, PS
Kidneys	4.03	54.52	25.89 ± 15.63	58.63 ± 16.5	fragment, fiber, pellet	PP, PP fiber, PLA
Spleen	1.52	12.24	5.42 ± 2.58	12.75 ± 7.09	fragment	PP
Small intestine	7.91	170.62	35.13 ± 33.93	61.06 ± 25.25	pellet, fragment, fiber	PP, PP fiber
Brain	12.00	127.48	36.21 ± 25.57	48.54 ± 11.20	fragment, fiber	PBAT, PE, PP, PP fiber
Lungs	8.70	159.71	56.80 ± 57.70	30.01 ± 9.49	fragment, fiber, long fragment	PP, PBAT

Figure 2B illustrates the size distribution of MPs across the organ tissue samples. In all human tissue samples, MP particle sizes ranged from 1.52 μm to 170.62 μm with an average size of 34.30 ± 34.91 μm. Notably, the largest MPs were detected in lung tissues (56.80 ± 57.70 μm), whereas the smallest occurred in the liver (5.13 ± 2.19 μm) and spleen (5.42 ± 2.58 μm). The size ranges of MPs in the seven major organs are detailed in Table 1. Figure 2C displays the proportion of different-sized MPs across organs. MPs with a diameter ≤5 μm were exclusively detected in the liver, spleen, and kidneys. The liver (100%) and spleen (91.67%) predominantly contained MPs smaller than 10 μm in diameter. MPs exhibited an inverse relationship between abundance and size. Approximately 59.9% of MPs were smaller than 20 μm, and more than 93.7% were smaller than 100 μm.

As shown in Figure 2D and Table 1, 9 types of MPs were detected in the human organ samples: PP (59.46%), poly(butyleneadipate-co-terephthalate) (PBAT; 11.71%), PE (9.00%), PP fibers (9.91%), PS (4.50%), polylactic acid (PLA; 2.70%), polycarbonate (PC; 0.90%), PE terephthalate (PET; 0.90%), and polyphenylene oxide (PPO; 0.90%). Analysis revealed that the heart samples contained the highest diversity of MP types, with seven distinct compositions identified, followed by the brain (four types) and kidneys (three types). Only one MP type (PP) was detected in the spleen, whereas two types were observed in the liver, small intestine, and lungs. PP was found in all seven organs, whereas PP fibers were exclusively detected in the liver and kidneys.

Figure 2E illustrates the five morphological categories of MPs identified in the human organ samples: fragments, elongated fragments, fibers, spheres, and granules. Fragment-shaped MPs predominated across all seven organs. Notably, only fragments were detected in the spleen. Fibrous MPs were present in all organs except the liver and spleen. Granular MPs were found exclusively in the liver, kidneys, and small intestine.

Validation of MP organ distribution via animal models

The animal model results revealed fluorescent signals in all organs across both the low- and high-concentration groups. Notably, stronger fluorescent signals were observed in the kidneys, small intestine, and liver, followed by the brain and lungs. In contrast, weaker signals were detected in the heart and spleen (Figures 3B and 3C). The organ-specific MP content closely mirrored findings in human tissues, confirming that this distribution pattern may apply universally to mammals.

Figure 3.

Accumulation of G-PS-MP beads in seven organs of rat

(A) Fluorescent MPs in seven organs.

(B) Fluorescence intensity in seven organs of low-dose MP-exposed rats.

(C) Fluorescence intensity in seven organs of high-dose MP-exposed rats.

The spatial distribution of these fluorescent signals is detailed in Figure 3A. In lung tissues, signals were predominantly localized near alveolar epithelial cells. In the liver, intense signals surrounded the central veins of hepatic lobules, diminishing radially outward. The small intestine exhibited prominent fluorescence in intestinal epithelial cells. In renal tissues, fluorescent signals were most evident near the glomerular basement membrane.

Raman image-based characterization of MP features in organs

Based on the MP segmentation results, we extracted 120-dimensional imaging features from the Raman images, comprehensively covering multiple attributes of the MPs. These high-dimensional features capture in detail the diverse morphological and structural characteristics exhibited by the MPs across different organs. Color-coded curves representing individual organs were superimposed to generate the integrated feature distribution plots shown in Figure 4 to visualize the distribution patterns of the MPs among organs. The two plots display Z score-normalized features adjusted to zero-mean and unit standard deviation, respectively. Variations in curve morphology, peak positions, and distribution widths reflect intrinsic differences in MP properties (e.g., morphology, grayscale intensity, texture, and color) across organs, providing a visual foundation for identification.

Figure 4.

Distribution map of imaging features of MP images of seven different organs

(A) Comparison of the mean values of various types of features.

(B) Comparison of standard deviation of various characteristics.

Analysis of MP Raman imaging features across seven organs

Through analysis of Raman imaging features of MPs in seven distinct organs, significant variations and partial commonalities in feature manifestations were identified. These differences reflect both organ-specific functional properties and migration patterns of MPs (as illustrated in Figure 5). Key findings include the following. In the lungs, MP particles exhibited larger sizes, predominantly elongated shapes, higher entropy and energy values, and complex textures. In the small intestine, MPs showed a disordered distribution, low particle density, intricate textures, and high heterogeneity. In the liver, MPs displayed uneven pixel intensities, distinct edges, high saturation levels, and significant intensity variations. In the kidneys, uniformly distributed low-density regions with relatively homogeneous texture patterns were observed. In the spleen, MPs showed a predominance of smooth areas, singular structural patterns, high intra-regional similarity, and uniform textures. In the heart, MPs exhibited elevated hue and saturation values, stark brightness contrasts, and localized hyperintense regions. In the brain, MP particles demonstrated high intensity values, smooth structures, and minimal grayscale variations between adjacent pixels. These differentiated characteristics provide robust evidence for analyzing the distribution patterns of MPs across various organs.

Figure 5.

Characteristic manifestations of MPs in seven different organs

Classification model for MPs in seven organs

This study adopted a combined approach integrating isometric mapping (Isomap) dimensionality reduction with a random forest algorithm, achieving a classification accuracy of 72.73%. Multiple comparative experiments were conducted to validate the effectiveness of the proposed method. Specifically, dimensionality reduction techniques, including principal-component analysis, logistic regression, and t-SNE, were paired with classification algorithms such as support vector machine and KNN to construct various model combinations for benchmarking. As shown in Figure 6, the Isomap-random forest hybrid method demonstrated superior performance in the MP organ attribution task. These results further confirm the advantages of the selected methodology for evaluating high-dimensional nonlinear data and its efficacy in capturing the discriminative distribution features of MPs across different organs.

Figure 6.

Classification results of MP organ attribution using different methods

Prediction of MP organ distribution

We collected 32 MP samples from air, rivers, and soil and classified their Raman images into predictions across seven specific organs. The results are illustrated in Figure 7. Each row in the figure represents the distribution probabilities of a sample among the seven organs, where higher peak values indicate greater likelihood of the sample accumulating in the corresponding organ. Through Raman spectral analysis of environmental MPs and machine learning modeling, we effectively predicted the organ-specific distribution patterns of MPs. This study provides new insights into the migration and accumulation patterns of MPs within organisms, offering a framework for assessing their potential biological impacts and health risks.

Figure 7.

Raman imaging test results of MPs in the environmental group

Discussion

This study systematically revealed, for the first time, the distribution patterns, organ-specific characteristics, and physiological correlations of MPs across seven human organs through integrated multi-organ sample analysis, animal model validation, and AI-driven imaging methodologies. The following discussion addresses the mechanisms of MP distribution, organ-specific heterogeneity, value of AI applications, and study limitations.

Mechanisms of cross-organ MP distribution: Exposure routes, blood flow dependency, and barrier penetration

As the primary exposure routes, the gut and respiratory tract exhibit distinct mechanisms of MP invasion. The gastrointestinal tract, acting as the “first gateway” for direct environmental MP contact, absorbs MPs via food chain sources (e.g., seafood²⁰ and plastic-packaged foods²¹) through small intestinal epithelial uptake. In this study, small intestinal MPs were predominantly fragments and fibers (59.46%) with particle sizes mostly between 20 and 50 μm, suggesting entry into the circulation via endocytosis or mechanical disruption of the intestinal barrier. Notably, particles <10 μm more readily penetrate tight junctions (TJs),²² whereas larger particles may enter the bloodstream indirectly through intestinal inflammation or dysfunctional mucosal layers.²³ The distribution characteristics of MPs in lung tissues revealed an alternative exposure pathway. The largest MPs detected in lung tissues (56.80 ± 57.70 μm), primarily PBAT fibers and PP elongated fragments, likely originated from airborne suspended particles (e.g., synthetic textile wear debris). However, the respiratory mucociliary system clears particles >5 μm; submicron particles (<1 μm) may deposit in alveoli and enter the circulation via macrophage phagocytosis.²⁴

This study confirms not only the widespread presence of MPs across human organs but also a significant correlation between MP abundance and organ blood supply levels. The liver, kidneys, and small intestine—organs with high blood flow (25%, 20%–25%, and regionally high perfusion rates of cardiac output, respectively)—exhibited the highest MP abundance (65.28 ± 23.94, 58.63 ± 16.50, and 61.06 ± 25.25 particles/g, respectively). These findings align with animal experiments. In rat models, G-PS-MPs showed significantly stronger signals in the liver, kidneys, and small intestine than in other organs, suggesting that MP distribution is highly dependent on dynamic circulatory transport. The brain demonstrated a notably high MP abundance (48.54 ± 11.20 particles/g), exceeding levels in the lungs and heart, with a predominance of sub-10-μm particles (91%). This result correlates with both the brain’s high blood flow (20% of systemic circulation), potentially accelerating MP accumulation, and the hypothesis that MPs penetrate the blood-brain barrier via the circulation. Recent studies have indicated that MPs induce oxidative stress in cerebrovascular endothelial cells,²⁵ increasing barrier permeability and ultimately promoting particle deposition.²⁶ Furthermore, this study provides new perspectives on potential links to neurodegenerative diseases,²⁷ such as Alzheimer’s disease.

Organ-specific MP characteristics: Functional and toxicological correlations

The physicochemical properties of MPs (size, shape, and chemical composition) exhibit significant variations across organs, indicating that their distribution is closely linked to organ function and microenvironment.

Lungs

MPs in lung tissues predominantly featured a larger size, elongated shape, high entropy and energy value, and complex texture. The larger dimensions and elongated morphology may relate to inhalation pathways, where particles undergo size-dependent deposition in the respiratory tract with larger particles retained in airways or alveoli. Notably, sharp-edged fragments detected in lung tissues (Figure 1) could directly damage alveolar epithelium, triggering localized inflammation and fibrosis.²⁸ These characteristics suggest that pulmonary MP toxicity arises from both physical airway obstruction and direct alveolar injury.

Small intestine

The average particle size of MP in the small intestine was 35.13 ± 33.93 μm, with particles ≤10 μm accounting for over 90%. This size range is highly consistent with the endocytic capacity of intestinal epithelial cells. Smaller particles may enter the circulation through “gaps” in intestinal mucosal TJs or via inflammation-mediated TJ relaxation, while larger particles mainly remain in the intestinal lumen. MPs in intestinal tissues were overwhelmingly composed of PP, with disordered spatial distribution and low particle density. PP is commonly used in food packaging²⁹ (e.g., cling film and beverage bottles) and textiles³⁰ (e.g., synthetic fibers). The enrichment of PP in the small intestine may be correlated with dietary exposure, but it may also be linked to the interactions between PP MP and the intestinal microenvironment. PP has a density of 0.9 g/cm,^3,31 which is lower than the density of intestinal fluid (approximately 1.0 g/cm³), causing it to remain in a suspended state within the intestinal lumen. This suspended state increases the probability of trans-epithelial transport. In contrast, high-density MPs may be rapidly eliminated due to gravitational sedimentation.³²

Liver

The liver exhibited the highest MP abundance and was also dominated by PP. As the systemic detoxification hub, approximately 90% of the hepatic blood supply originates from the intestines via the portal vein,³³ allowing intestinal MPs to accumulate preferentially. Hydrophobic PP MPs resist biliary excretion, potentially accumulating in hepatocytes and obstructing bile canaliculi. High saturation levels and pronounced intensity variations in imaging features suggest interactions with metabolites or toxicants.

Kidneys

Renal MPs showed relatively uniform size and composition, with imaging features revealing homogenously distributed low-density regions and consistent texture patterns. The uniform texture may reflect the kidneys’ efficient filtration and transport mechanisms. Although balanced distribution implies limited acute physical impact, chronic MP deposition could progressively impair tubular function.³⁴

Spleen

Spleen MPs were characterized by low abundance, small sizes, and high homogeneity, aligning with imaging findings of smooth structural patterns, minimal intra-regional variation, and uniform textures. This likely results from rigorous circulatory filtration prior to splenic entry, where only small, rounded particles persist. Additionally, robust phagocytic activity by splenic macrophages contributes to reduced MP retention. As a vital component of the lymphatic system, the spleen has been confirmed in multiple studies to detect MPs after different exposure routes. Shanmugiah et al. found that MP signals were detected in both the spleen and thymus of mice after inhalation exposure to model MP.³⁵ Wu et al. revealed that, through gavage-established MP exposure models, the strongest MP fluorescence signals were observed in the intestine and spleen.³⁶ Peyer’s patches, the core structure of intestinal mucosal immunity, are covered by M cells that actively take up intestinal particles and antigens; studies have shown that small MP particles (<5 μm) can cross the intestinal barrier via M cells and enter the lymphatic system from the intestine. These findings suggest that the lymphatic system may play a critical role in the distribution process of MPs.

Heart

Cardiac MPs displayed average abundance, size, and morphological uniformity across organs. As the central circulatory hub, the heart evenly redistributes bloodborne MPs to all organs while retaining a proportional baseline level.³⁷

Brain

MPs in brain tissues exhibited high intensity values, smooth structures, and minimal grayscale variations between adjacent pixels. Statistically, 70% of cerebral MPs fell within the 20- to 50-μm range, indicating size-selective deposition. This uniformity may arise from the selective filtration of the blood-brain barrier, which permits entry only to MPs within specific size thresholds. The predominance of smooth, high-intensity particles further supports adaptation to the brain’s unique biophysical environment.

AI-driven MP research: Methodological breakthroughs and applications

In previous studies on MP detection and characterization in humans, MP properties were primarily described by shape, size, and other basic attributes.^{6,7,8,9,10,11,38,39,40,41,42} However, these features exhibit multidimensional and nonlinear relationships. Machine learning overcomes the limitations of traditional statistical methods by algorithmically uncovering complex associations. This study pioneered the use of deep learning to extract a 120-dimensional imaging feature set from Raman spectroscopic images of MPs via convolutional neural networks, enabling unprecedented discrimination of MP variations. Building on these high-dimensional features, we developed a predictive model to infer MP distribution across organs based on their intrinsic characteristics. This methodological advance holds transformative potential. (1)Toxicity mechanism elucidation: predicting deposition sites facilitates investigation of MP-induced localized inflammation (e.g., hepatic fibrosis) or systemic effects (e.g., immune responses), clarifying pathogenic pathways and toxicological mechanisms across organs. (2) Exposure-risk stratification: by identifying MP accumulation hotspots and integrating epidemiological data, this approach enables analysis of health risk correlations in high-exposure populations,^43,44 guiding early screening protocols and tiered protective guidelines for disease prevention. Through interdisciplinary collaboration, this research framework will advance comprehensive assessments of the long-term health impacts of MPs, inform public health policymaking, and ultimately contribute to reducing MP exposure and safeguarding human health.

Conclusion

Using an interdisciplinary approach, this study systematically elucidated the organ-specific distribution patterns and heterogeneous health risks of MPs in humans, establishing the first AI-driven predictive model for MP biodistribution. These findings not only provide a novel paradigm for MP toxicology research but also establish a scientific foundation for targeted intervention strategies (e.g., prioritized regulation of PP plastics) and public health policymaking. As global plastic pollution intensifies, deciphering the “in vivo journey” of MPs will be critical for safeguarding human health.

This study has several limitations. First, although it offers critical evidence of clarifying MP absorption-transport pathways in humans, further investigation is required to explore potential transport mechanisms via lymphatic and interstitial fluids. Additionally, the observed MP compositions may reflect individual variability and geographical biases, as the postmortem donors were predominantly from Northwestern China, and the sample size was limited. Future studies should expand specimen collection to enhance both sample diversity and geographical representation. Third, although combined toxic effects represent an important research area, the synergistic effects between MPs and environmental pollutants have not been explored here. Future research should incorporate more complex analytical techniques to comprehensively assess the health risks of MP exposure.

Resource availability

Materials availability

No material was generated in the present study.

Data and code availability

Data and code are available from the corresponding author upon reasonable request.

Funding and acknowledgments

This study was supported by the National Natural Science Foundation of China (82271913 and 42277207), the Shaanxi Province Key Industrial Innovation Chain under grant 2023-ZDLSF-12, the Shaanxi Provincial Natural Science Foundation General Project-General Program (2023-JC-YB-751), the FMMU special research project of cross-cooperation (2024JC044), and the Fundamental Research Funds for the Central Universities (GK202401003). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Author contributions

Z.C., conceptualization, data curation, funding acquisition, writing – original draft, and writing – review & editing; Y.L., data curation, methodology, resources, and validation; Q.Y., data curation, formal analysis, writing – original draft, and writing – review & editing. A.L. and S.G., formal analysis, methodology, software, validation, and visualization. L.Z., data curation, methodology, and validation; H.L. and Y.W., investigation, project administration, and writing – review & editing; T.D., conceptualization, funding acquisition, methodology, supervision, validation, and writing – review & editing. All authors contributed to the manuscript and approved the final version.

Declaration of interests

The authors declare no competing interests.

Published Online: July 4, 2025