Anatomy-resolved digital twin framework for personalized light dosimetry in lung therapies

Significance

Understanding light propagation and energy deposition in biological tissues is essential for advancing photodynamic and other light-based therapies. Here, we demonstrate a patient-specific computational framework that integrates Computer Tomography (CT) derived 3-Dimensional (3D) anatomical models, tissue-specific optical properties, and Graphics Processing Units (GPU) accelerated Monte Carlo simulations to predict light propagation, absorbed energy, and fluence rate distributions within target organs. This approach enables personalized treatment planning, supports predictive optimization of therapeutic parameters prior to intervention, and represents a step toward clinically translatable digital twin-guided precision medicine for light-based therapies.

Abstract

Light therapies, such as photobiomodulation and photodynamic therapy, have demonstrated effective therapeutical response. However, external irradiation and targeting internal organs is challenging due to high light attenuation and the interindividual anatomical variability, strongly affecting light dosimetry. When considering lung diseases, such as pneumonia and inflammatory conditions, light dosimetry using transcutaneous delivery must consider the total light path from skin coupling through thoracic tissues to the fluence rate reaching the lungs. We here present a computational model based on Monte Carlo simulation developed in house to determine the light dosimetry in the lungs for different wavelengths in normal and diseased lungs. The simulation uses the actual anatomical models derived from human CT scans to generate 3D representations of the thoracic cavity, including 12 distinct tissue types. The lungs were segmented into normal, collapsed regions, infiltrate, and blood vessels. Wavelength-dependent absorption and scattering coefficients for each tissue type were estimated by the reported data in the literature. Our analysis focused on normalized energy deposition profiles defined as µ_a*fluence rate and fluence rate distributions within each lobe from normal, focal, and diffuse pneumonia, and COVID-19 conditions, under different irradiation positions. The results reveal significant variations in energy deposition and fluence rate distribution across different lung conditions and irradiation parameters, highlighting the need for individualized light dosimetry planning.

Lung infections, particularly pneumonia, are a critical global health issue and the third leading cause of death worldwide (1). pneumonia affect all age groups but are most severe in children, the elderly, and chronically ill patients, especially those in intensive care (2–4). The urgency of addressing these infections is heightened by the rise of antimicrobial resistance (AMR) and the emergence of new pathogens (5). The World Health Organization (WHO) has projected that, without effective interventions, AMR can lead to 10 million annual deaths by 2050 (6). Even before the COVID-19 pandemic, concerns about AMR and its economic impact were significant, with the WHO warning of the potential for a global health crisis (6). Current pneumonia treatments include antibiotics, antivirals, corticosteroids, and supportive interventions such as oxygen therapy and mechanical ventilation. Despite widespread use, these approaches face major limitations, including rising AMR, adverse effects, and drug-delivery challenges.

Given these limitations, photonic therapies have emerged as treatment approaches for diverse pathologies targeting endogenous or exogenous photosensitive molecules to induce therapeutic tissue response (7–10). Photonic techniques include, photobiomodulation (PBM) (11, 12), laser ablation (13, 14) and photodynamic therapy (PDT) applied to modulate inflammation (15) enhance healing (16), resect abnormal tissues, and locally treat cancer (17, 18) and infections. The diversity of responses arises from distinct light–tissue interaction patterns determined by irradiation parameters wavelength, fluence rate, interaction time, and fluence. In this sense, light dosimetry, which is related to the irradiation parameters to be used in each application, is essential to achieve the desired. In the context of lung infections and inflammation, antimicrobial PDT and PBM are of particular interest.

Antimicrobial photodynamic therapy (aPDT) emerges as a promising method due to its noninductive nature of resistance (19). aPDT uses a photosensitizer (PS), light at a specific wavelength, and molecular oxygen to generate reactive oxygen species (ROS) that induce cell death via oxidative stress (20, 21). Unlike conventional antimicrobials, aPDT targets multiple biomolecular sites through ROS-mediated damage, markedly reducing the likelihood of resistance development (4, 5, 22). Historically applied to tumors and localized infections (23, 24), aPDT’s potential against pneumonia using external illumination requires overcoming light attenuation through the thoracic wall and determining whether sufficient photon flux reaches the lungs to achieve pathogen inactivation (25).

PBM has gained attention as a potential nonpharmacological treatment to counteract COVID-19-induced inflammation (26). The COVID-19 pandemic has overwhelmed global healthcare systems, affecting millions of patients. Beyond respiratory issues, the SARS-Cov-2 virus triggers a systemic inflammatory response, leading to long-term health problems (27). A key factor in severe COVID-19 cases is the “cytokine storm,” in which excessive proinflammatory cytokines drive disease progression and mortality (28). PBM uses low-intensity red and near-infrared light to regulate cellular processes and inflammatory pathways (26, 28). By targeting these mechanisms, PBM may complement existing treatments and improve clinical outcomes. As research continues to unravel the complexities of SARS-CoV-2, innovative therapies like PBM may offer new solutions for managing the disease’s systemic effects (28).

Despite the potential of aPDT and PBM, treating internal organs such as the lungs remains challenging. Bronchoscopic light delivery is invasive and generally limited to mechanically ventilated patients; thus, external illumination through the thoracic wall represents a more attractive noninvasive approach. Light must travel through multiple tissue layers in the thoracic wall, including skin, fat, muscle, and bone, before reaching the thoracic cavity and the lungs. The optical properties related to the heterogeneity of lung tissues add complexity for the efficacy planning of light dosimetry. Understanding light transport in the presence of focal and diffuse pneumonia and COVID-19–related alterations is therefore essential for protocol optimization. To address this clinical need, we employed our in-house Monte Carlo (MC) framework to investigate light delivery through the thoracic wall and fluence rate distribution within the lungs, considering distinct pathological conditions and tissue architecture of each volunteer. Using detailed anatomical models reconstructed from human computed tomography (CT) scans, we created 3D voxelized representations of the thorax, incorporating 12 distinct tissue types: skin, subcutaneous fat, muscle, bones, heart, liver, stomach, kidney, and in the normal, collapsed, infiltrate, and blood vessels lungs. These models allowed for accurate full 3D volumetric simulation of light transport and energy deposition. The used optical properties library, based on comprehensive reviews (14, 29–32), provided the necessary absorption and scattering coefficients for each tissue type across a range of wavelengths.

This study normalized energy deposition profiles and fluence rate distributions within specific lung lobes (left lower lung LLL, left upper lung LUL, right upper lung RUL, right middle lung RML, and right lower lung RLL) under different light positioning configurations and lung conditions of normal, focal, and diffuse pneumonias, and COVID-19-inflamed lungs. We propose a MC framework to model light propagation based on patient-specific anatomy, generating absorbed energy maps and fluence rate variability across disease states. This in silico dosimetry platform enables individualized irradiation planning according to wavelength, lung condition, and affected regions.

Results

To meet the simulation objectives, the initial step consisted of modeling lung anatomy and its pathological variations. CT-derived DICOM volumes were processed to generate tissue-specific representations. As shown in Fig. 1A, a 3D array was extracted from Digital Imaging and Communications in Medicine (DICOM) data, with a lung window setting applied to enhance contrast for optimal visualization. Using 3D Slicer (33), segmentation masks are generated for various tissue types based on intensity thresholds for specific Hounsfield unit ranges (34). These masks were converted into STL surface models representing skin, bones, lungs, and thoracic and abdominal organs. The STL models were subsequently voxelized (0.1 × 0.1 × 0.1 cm³) to produce computational grids suitable for optical simulations while preserving spatial resolution.

Fig. 1.

Pipeline and results of image segmentation, 3D reconstruction, and optical properties analysis. (A) Set of DICOM images representing the CT scans which are then subjected to image segmentation and 3D reconstruction of a STL model for each tissue type, followed by voxelization. (B) 3D reconstructions for four patients (control volunteer, focal pneumonia, diffuse pneumonia, and COVID-19 patients). (C) Voxelization of the STL model obtained from the segmentation for lung, collapsed lung tissue, blood vessels, and infiltrates for control volunteer and diseased conditions (focal pneumonia, diffuse pneumonia, and COVID-19). (D) Spectral dependence of absorption (μ_a) and reduced scattering (μ_s′) coefficients for different tissue types across a range of wavelengths from 600 to 1,000 nm.

Fig. 1B shows 3D reconstructions of four cases: a healthy volunteer, focal pneumonia, diffuse pneumonia, and COVID-19. The diseased lungs exhibit altered tissue distribution and density, with structural changes consistent with inflammation and collapse. Compared to the healthy control, these models illustrate the extent and localization of pathological remodeling.

Fig. 1C shows the voxelization of segmented STL lung models, including normal lung tissue, collapsed regions (atelectasis, where alveoli lose air), blood vessels, and infiltrate (indicating inflammation), across four cases: healthy control lungs, focal pneumonia, diffuse pneumonia, and COVID-19. The voxelized models quantify the volumetric fractions of each tissue type. Healthy lungs exhibit a higher percentage of normal lung tissue (91.11%) compared to focal pneumonia (58.93%), diffuse pneumonia (54.14%), and COVID-19 (69.98%) accompanied by increased volumes of collapse, infiltrate, and vascular components in diseased states. These quantitative differences provide structural metrics for characterizing disease severity and regional involvement. Fig. 1D presents the wavelength-dependent absorption and reduced scattering coefficients of the analyzed tissues obtained from literature (29, 35). Sandell and Zhu reported substantial variability in tissue optical properties across the literature (35), in this study we adopted the optical property framework described by Steve Jacques (29). In this model, tissue absorption is parameterized as a function of the relative contributions of the main tissue chromophores, while scattering is described in terms of the reduced scattering coefficient and scattering power dependence. Absorption spectra show chromophore-specific features, with oxy- and deoxy-hemoglobin dominating in the visible range (~750 nm) and water absorption increasing in the near-infrared (900 to 1,000 nm). Reduced scattering coefficients, modeled using a power-law equation (Eq. 2), exhibit distinct tissue-specific behaviors linked to structural composition and density. These optical properties critically influence light penetration depth and energy distribution, making them essential for accurate light dosimetry. By integrating these optical characteristics, the MC simulations offer a robust framework for predicting light propagation within the target tissue to individualize and improve light dosimetry planning.

SI Appendix, Fig. S1 illustrates the 3D segmentation workflow and anatomical representation of the thoracic region. The upper panel shows the reconstructed STL model of a control volunteer, including major tissues. The lower panel present animated 3D rendering highlighting lung segmentation across clinical conditions, demonstrating disease-specific morphological and volumetric differences.

Comparative Analysis of Wavelengths, Energy Absorption, and Illumination Configurations.

To evaluate how light energy is deposited in different lung regions, MC simulations were performed to model absorbed energy distribution across the thoracic cavity for distinct wavelength and positioning configurations of the light sources; this analysis was performed only for the control volunteer. results are shown in Fig. 2. Fig. 2A presents absorbed energy distributions, defined as ${\mu }_a\mathrm{\Phi }$ (W/cm³), for four illumination configurations. Columns correspond to individual configurations, while rows show external thoracic views (Top) and lung surface distributions (Bottom). Absorbed energy was computed as the product of simulated irradiance (W/cm²) and tissue-specific absorption coefficients and displayed on a logarithmic scale. Warmer colors (−30 to 5, log units) indicate higher energy absorption. These spatial distributions reveal how each illumination configuration directs energy within the thoracic cavity, highlighting potential coverage areas and the desired regions where the light dose needs to be delivered to. The configurations exhibited significant differences in their energy deposition patterns, underscoring the importance of laser positioning configuration in achieving effective therapeutic targeting of specific lung regions and highlighting its importance for tailoring photonic treatments to anatomical features and disease localization.

Fig. 2.

Overall energy deposition. (A) Sagittal, coronal, and axial view of the absorbed energy (log scale) of the photons path for 808 nm wavelength for the four different light sources setups. (B) Overall energy deposition in the whole lungs for different wavelengths (660 nm, 750 nm, 850 nm, and 950 nm) and laser positions (setup 1, setup 2, setup 3, setup 4). (C) Overall absorbed energy deposition for 808 nm in individual lung lobes (LLL, LUL, RLL, RML, RUL) for different laser positions.

Fig. 2B quantifies total lung energy deposition for wavelengths of 660, 750, 808, 850, and 950 across four illumination setups. Energy deposition is maximal at 808 nm in all configurations, indicating more favorable tissue interaction at this wavelength. Among the setups, configuration 4 yields the highest deposition (~10% of the incident energy at the skin), likely due to reduced attenuation when illuminating the lateral and dorsal thoracic walls rather than the anterior chest. In contrast, frontal illumination results in lower lung energy delivery, primarily due to rib attenuation and increased adipose tissue in the breast region, which limits photon penetration.

Fig. 2C further discriminates the energy deposition by individual lung lobes [left lower lobe (LLL), left upper lobe (LUL), right lower lobe (RLL), right middle lobe (RML), and right upper lobe (RUL)] at 808 nm for each illumination configuration. The data reveal that the lower lobes, except for setup 1, usually receive higher energy deposition across most setups, a relevant observation given that these lobes are often affected by bacterial pneumonia due to gravity-driven accumulation of fluids and pathogens in the lower parts of the lungs, making them more susceptible to infection (36, 37). Understanding lobe-specific energy absorption is crucial for tailored therapies, especially in treating infections like pneumonia that may preferentially impact lower lobes while remaining adaptable for infections affecting other lung regions. At the same time, pneumonia may affect any lung region, including upper and multiple lobes depending on etiology and disease progression (3, 4, 36, 38).

SI Appendix, Fig. S3 shows the spatial light dose distribution for external illumination at 100 mW/cm² applied for 80 min, computed using patient-specific anatomical segmentation. The resulting 3D dose map (J/cm²) decreases with tissue depth due to absorption and scattering, with higher exposure near the lung surface and rapid attenuation in deeper parenchyma. Regional dose values can be extracted to estimate PDT thresholds within specific anatomical regions while accounting for the optical and photochemical properties of the selected PS. These simulations support prediction of effective treatment zones and personalized PDT planning based on patient-specific geometry and optical properties.

Absorbed Energy and Normalized Fluence Rate across Different Disease Conditions.

Fig. 3 presents boxplot distributions of normalized fluence rate by lung lobe for each illumination configuration, illustrating intralobar heterogeneity. In setup 1, fluence rate is uneven across lobes, with lower values in LLL and RLL, correlating with their higher energy deposition observed with the anterior and posterior panels. Setup 2 exhibits greater variability, particularly in the RUL, indicating less uniform lateral illumination. Setup 3 shows relatively homogeneous fluence rate across lobes, with slightly higher values in LUL, RML, and RUL. Setup 4 yields the highest fluence rate overall, especially in LLL and RLL, reflecting the combined lateral, anterior, and posterior illumination geometry. Among the tested configurations, setup 4 provides the most uniform distribution across lung regions and may therefore represent the most favorable configuration for whole-lung phototherapy. Alternative geometries may be preferable when disease is confined to specific lobes.

Fig. 3.

The boxplots for the fluence rate at 808 nm and the four illumination setups: Boxplot distribution of normalized fluence rate (W/cm²/W_delivered) at 808 nm for each lung lobe (LLL, LUL, RLL, RML, RUL) under various laser configurations: setup 1, setup 2, setup 3, setup 4. The sketch diagrams next to each boxplot indicate the laser positions on the patient’s body.

The left panels of Fig. 4 show absorbed energy deposition by lung lobe for focal pneumonia (Fig. 4A), diffuse pneumonia (Fig. 4B), and COVID-19 (Fig. 4C) under 808 nm irradiation using setup 4. Each condition exhibits a distinct deposition pattern. In the focal pneumonia, collapsed regions exhibit the highest absorbed energy deposition, followed by healthy tissue in the left lobes. Blood vessels also present a relatively high absorbed energy, mostly due to the high concentration of this structure near the right middle and lower lobes, as it can be observed in the 3D of the vessel tissue plot. In diffuse pneumonia, collapsed regions again dominate, but absorption is more uniformly distributed across lobes and vascular structures due to the widespread nature of the disease. In COVID-19 case, energy deposition pattern remains elevated in collapsed regions, while infiltrate and vascular compartments exhibit comparatively lower absorption, reflecting disease-specific structural alterations. The right panels present boxplot distributions of normalized fluence rate for each lung lobe under the three pathological conditions.

Fig. 4.

808 nm and setup 4 illumination: Overall energy deposition (%) in the lungs, collapsed regions, infiltrate, and blood vessels. Boxplot of the normalized fluence rate distribution for each lung lobe (LLL, LUL, RLL, RML, RUL) in the presence of (A) focal pneumonia, (B) diffuse pneumonia, and (C) COVID-19.

These plots characterize intralobar fluence rate variability. In focal pneumonia, fluence rate is relatively consistent across lobes, with moderate median values, although increased dispersion is observed in LUL and RUL, likely reflecting localized pathology. Diffuse pneumonia exhibits broader variability across all lobes, with greater divergence between mean and median values, particularly in LUL and RML, indicating heterogeneous light penetration. In COVID-19, fluence rate variability is more pronounced, with notably reduced values in the RUL compared to other lobes, consistent with the nature of COVID-19 pathology, where certain regions of the lung are more affected than others.

Experimental and Simulated Irradiance Distributions.

cross-validation was performed with two volunteers. SI Appendix, Fig. S4. illustrates the simulation workflow for Subjects 1 and 2, where MC light transport models were generated from anatomical segmentations and experimental illumination configurations. Two illumination panels (each delivering 100 mW/cm²) were positioned, one on the left lateral and the other on the left posterior thoracic surface to simulate simultaneous exposure. Fluence rate and absorbed energy distributions were computed across tissue layers, including skin, lung parenchyma, and internal organs. By isolating lung and skin domains, absorbed dose could be quantitatively analyzed, enabling assessment of intersubject variability in light delivery and tissue penetration under identical external illumination conditions.

SI Appendix, Fig. S4 presents fluence rate distributions and statistical characterization for Subjects 1 and 2 under dual-panel illumination (2 × 100 mW/cm²). The upper panels show 3D fluence rate maps of the whole body, isolated lungs, and skin, illustrating light propagation through thoracic tissues. Histograms quantify the probability density of fluence rates within skin and lung domains, demonstrating exponential attenuation with depth. Boxplots summarize irradiance variability, revealing greater heterogeneity in superficial tissues compared to deeper pulmonary structures.

Fig. 5 compares experimentally measured irradiance (Fig. 5A) with MC simulations (Fig. 5B and C). Experimental skin irradiance maps (Fig. 5A), obtained through optical dosimetry, show a spatial gradient with higher values near the laser panel edges and progressive decay with distance. These data serve as validation for the in-silico 3D model. Simulated irradiance distributions (Fig. 5 B and C), computed using the panel’s initial irradiance, reproduce this spatial pattern, with peak values near the source and attenuation away from it. The 3D surface reconstructions demonstrate the influence of body curvature and surface orientation on light propagation. Fig. 5C presents a probability distribution of irradiance values at the skin surface, highlighting the likelihood of encountering specific fluence rate levels across the measured region. The logarithmic scale of the probability distribution emphasizes that most of the irradiance values fall within a lower intensity range, with higher irradiance values confined to direct exposure regions.

Fig. 5.

Experimental and MC simulated irradiance distributions on the skin. (A) Experimental irradiance distribution measured on the skin of two volunteers using optical dosimetry. The heatmaps show spatial variations in irradiance (mW/cm²), with higher intensities near the laser panel placement. (B) MC simulation results depicting the 3D irradiance distribution on the volunteers’ skin. The logarithmic color scale highlights regions of high and low fluence, incorporating the initial irradiance from the laser panels. (C) Probability distribution of irradiance values on the skin, derived from the MC simulations. The logarithmic scale shows the likelihood of encountering specific fluence values, revealing a higher frequency of lower irradiance levels due to light scattering and surface curvature.

To provide a surface-level cross-validation of the MC model, simulated fluence-rate values were extracted from an anterior thoracic ROI corresponding to the experimental optical dosimetry measurement area (SI Appendix, Fig. S5). Within this ROI, the simulated skin fluence-rate distribution exhibited substantial spatial heterogeneity, spanning several orders of magnitude due to differences in tissue curvature, source-detector geometry, and attenuation through thoracic tissues. Although the median simulated fluence rate remained considerably lower than the experimental measurements, the upper tail of the simulated distribution overlapped with the experimentally observed range. Specifically, the simulated ROI values extended up to approximately between 0.72 to 0.78 mW/cm² (100th percentile), which is within the same order of magnitude as the measured anterior thoracic fluence rates. These findings indicate that the model predicts detectable light propagation to regions distant from the illumination panels and reproduces the experimentally observed magnitude of transmitted light. However, because the simulation incorporates literature-based optical properties and includes substantial spatial variability across the ROI, the agreement should be interpreted as semiquantitative rather than as a direct point-by-point validation. A more rigorous validation of internal light propagation would require patient-specific optical property measurements, as well as a consideration of the melanin content, and invasive dosimetry, such as bronchoscopic measurements within the lung, which will be investigated in future clinical studies.

Comparative Analysis of Tissue-Specific Energy Deposition.

Fig. 6 compares tissue composition distribution a) and normalized energy deposition b) across various subject groups, including healthy volunteers and patients with distinct lung pathologies. In Fig. 6A, the proportion of healthy lung tissue is similar across groups, although modest reductions in diseased cases. Collapsed lung tissue remains relatively stable, but infiltrate regions increase significantly in pneumonia and COVID-19 groups. The blood vessel fraction varies slightly, with a marked increase in diffuse pneumonia and COVID-19 cases.

Fig. 6.

Quantitative comparison of tissue distribution and energy deposition across different subject groups. (A) Percentage of tissue composition for healthy lung tissue, collapsed lung tissue, infiltrate, and blood vessels in different subjects, including a control volunteer, two healthy subjects (Subject 1 and Subject 2), and patients with focal pneumonia, diffuse pneumonia, and COVID-19. (B) Normalized overall energy deposition (%) for each lung tissue type, illustrating how different pathological conditions affect light distribution and absorption. The logarithmic scale highlights variations across multiple orders of magnitude.

Fig. 6B shows the overall energy deposition, normalized for intersubject comparison. Healthy lung tissue accounts for the largest fraction of absorbed energy in all cases, with minor variations reflecting tissue heterogeneity. Infiltrate regions exhibit reduced absorption compared to normal parenchyma, particularly in focal and diffuse pneumonia. In COVID-19, a greater proportion of energy is deposited in vascular compartments, consistent with increased optical absorption in highly vascularized regions. Collapsed tissue displays a distribution like healthy lung, with modest deviations attributable to structural alterations.

Discussion

To optimize light therapies for lung diseases, we integrate clinical expertise with advanced computational modeling to evaluate anatomical and pathological variability in a patient-specific framework. Below, we describe the DICOM-based image segmentations and MC simulations, along with the adapted methods and their applications.

Patient Descriptions and Segmentation Analysis.

Control volunteer.

A female patient undergoing tuberculosis screening while receiving immunobiological therapy for lupus. Despite immunosuppression, CT imaging revealed only mild infiltrates, consistent with her clinical profile. Segmentation analysis highlighted predominantly central blood vessels with a slight infiltrate, consistent with near-normal lung morphology.

Focal pneumonia.

A female patient presenting significant pathology in the right lower lobe, characterized by prominent infiltrates, edema, and collapsed lung tissue. Increased vascularity suggested elevated perfusion to the affected region. While the primary infection was localized, milder signs of infection were evident in other lobes. Segmentation effectively differentiated infiltrate, atelectasis (collapsed tissue), and vascular congestion. This detailed modeling underscores the necessity of considering the entire lungs for treatment planning, as airway contamination often extends beyond localized infections.

Diffuse pneumonia.

A male patient with a case of severe, bilateral pathology predominantly affecting posterior lung segments, consistent with bronchial aspiration. widespread infiltration and collapse were present across lobes. Segmentation confirmed extensive involvement of posterior regions, providing a basis for comprehensive therapeutic planning.

COVID-19 pneumonia.

A male patient admitted for cardiac decompensation secondary to diarrhea, who developed mild viral pneumonia. Segmentation revealed diffuse, sparse infiltrates with a distinctive ground-glass appearance. Unlike bacterial pneumonia, vascular enlargement was not prominent, reflecting the absence of pronounced immune-mediated circulation changes. Greater anterior lobe involvement was observed compared with other cases, highlighting heterogeneity in disease distribution.

Method Adaptation and Segmentation for MC Simulations.

Optimizing light therapy for lung diseases requires strategic light source positioning to achieve uniform and targeted illumination while accounting for attenuation due to scattering, absorption, and shadowing. Patient-specific anatomical variability motivates the integration of CT-derived three-dimensional models with MC simulations to predict light propagation and absorption and guide illumination planning.

Central to this process is accurate segmentation of lung regions of interest (ROIs), enabling differentiation between healthy tissue and pathological features such as infiltrate, collapsed tissue, and blood vessels. In pneumonia and COVID-19, this distinction supports tailored strategies, from lobe-specific targeting in focal disease to broader coverage in diffuse involvement. CT-based voxelized representations provide anatomically faithful inputs for MC simulations, although their accuracy depends on reliable anatomical characterization (39–41). Quantifying light dose within selected ROIs is essential for evaluating therapeutic efficacy. Comparisons between normal and diseased lungs reveal variations in light propagation and absorption, offering insights into disease-specific therapeutic responses. As the study progresses, these methodologies contribute to refining PDT protocols and advancing understanding of light–tissue interactions in pulmonary applications.

MC simulation is a powerful tool for modeling the complex process of radiative energy transfer, which is crucial for understanding and optimizing lung therapies that utilize light-based protocols (42, 43). MC simulation is a statistical method that models the trajectories of millions of photons as they propagate, scatter, and are absorbed within biological tissues. In lung therapies, MC simulations enable the prediction of light distribution, fluence rate, and energy deposition by accounting for complex anatomy and tissue optical properties. These models support the optimization of light source placement and wavelength selection for PDT, improving treatment efficacy, dosimetry accuracy, and noninvasive optical monitoring of pulmonary tissues (44).

For PBM (45, 46) where, it is needed low-level irradiance values (12, 47, 48), but without the prediction of the fluence rate levels within the target organ, the determination of the exposure time to achieve the treatment dose is a reachable task. The comprehensive pipeline described in this study integrates advanced image processing and simulation techniques to provide a detailed analysis of tissue structures and their optical properties, offering critical insights for optimizing PDT and PBM protocols. These simulation-based approaches allow us to explore a wide range of parameters and scenarios, optimizing photonic protocols by allowing individualized illumination planning. This not only improves the chances of successful outcomes but also minimizes risks for patients.

Marked differences in tissue composition and optical properties between control and diseased states underscore the importance of individualized treatment planning in light-based therapy. Accurate segmentation and 3D reconstruction enable patient-specific dosimetry, while spectral absorption and scattering data support optimization of wavelength and irradiation parameters. The comprehensive analysis of energy deposition and fluence rate distribution demonstrates that both wavelength selection and source geometry critically influence light delivery and uniformity within lung tissue. The observed variability between different lobes further reinforces the importance of anatomy-informed planning. Among the tested configurations, a multiangle laser configuration, particularly 100L + 200B + 100R at 808 nm wavelength provided the most uniform and effective light delivery, maximizing energy deposition and fluence rate distribution. This could significantly enhance the efficacy of light therapies, improving therapeutic outcomes for patients with lung diseases.

Several modalities of photonic-based therapies have been proposed as a promising approach for treating lung diseases, with their effectiveness governed by the interplay of three key parameters: wavelength, fluence rate, and laser exposure time. Among these, wavelength determines tissue penetration. NIR light, particularly in the 780 to 810 nm range, has shown superior performance due to its deep tissue penetration and low absorption by biological components (49). In particular, 808 nm efficiently activates photosensitizers such as indocyanine green (ICG), attributed to enhanced photobleaching and efficient ROS generation (49, 50). Comparative studies confirm that 808 nm light penetrates more effectively than 780 nm through biological barriers, supporting its use in extracorporeal PDT for pneumonia and other thoracic infections (49, 51, 52). Irradiance (fluence rate, mW/cm²) modulates ROS production and microbial inactivation but must be balanced to prevent oxygen depletion and thermal effects. Lower fluence rates, such as 75 mW/cm², have shown improved efficacy compared with higher values due to reduced oxygen depletion (53, 54), whereas a minimum threshold of 8.6 mW/cm² is required for microbial inactivation (55). Exposure time determines total fluence, with clinical PDT typically requiring 15 to 45 J/cm² and preclinical models up to ~198 J/cm² (25, 56). Longer exposures at lower intensities help preserve tissue oxygenation and reduce thermal risk (53, 57). Importantly, these parameters are interdependent: deeper-penetrating wavelengths allow lower intensities and shorter exposure times; meanwhile, lower irradiance levels require longer exposure times to achieve adequate fluence without depleting oxygen. To ensure precise and uniform light delivery, advanced technologies such as real-time dosimetry and optical surface applicators have been developed (58–61). Overall, successful thoracic PDT or PBM depends on coordinated optimization of wavelength, irradiance, and exposure time according to target tissue distribution and pathology.

Energy Deposition and Fluence Rate Distribution across Different Lung Diseases.

The MC simulations of light propagation at 808 nm in segmented CT data from patients with focal pneumonia, diffuse pneumonia, and COVID-19 highlight how pathological alterations in lung tissue architecture directly modulate optical energy deposition and fluence rate with respect to the control volunteers. Collapsed lung regions consistently showed the highest energy deposition across all conditions (3 to 4.5%), a finding that reflects increased tissue density, reduced air content, and corresponding elevations in both absorption and scattering coefficients. In contrast, normal aerated parenchyma absorbed the least energy (~1 to 2%), establishing a baseline that underscores how the optical signature of infiltrated or consolidated regions diverges from healthy lung. Infiltrated tissues exhibited intermediate deposition (2 to 3%), with values tending higher in COVID-19, consistent with the protein-rich exudates and cellular infiltrates known to alter refractive index and enhance absorption in inflamed regions. Disease-specific patterns were also evident: focal pneumonia showed sharply localized zones of high deposition with strong boundaries between affected and healthy parenchyma, diffuse pneumonia demonstrated more homogeneous but elevated deposition across the parenchyma, and COVID-19 displayed a unique bilateral and peripheral signature, with ground-glass opacities producing intermediate scattering zones.

Fluence rate analyses further revealed how regional anatomy shapes optical behavior under pathological conditions. The lower lobes, particularly LLL and RLL, demonstrated the greatest variability and lower median fluence rate in COVID-19 cases, aligning with gravitational pooling of inflammatory fluid, while upper lobes maintained more consistent lower fluence rate values across diseased conditions. These spatial differences emphasize the importance of regional anatomy in light distribution and the need for targeted treatment planning. Vascular structures consistently showed minimal energy deposition (~1 to 2%), suggesting that despite profound parenchymal remodeling, vascular optical properties remain relatively preserved. Collectively, these results not only provide a mechanistic explanation for the optical signatures of pneumonia and COVID-19 but also carry translational implications of photonic techniques.

Additionally, the analysis of energy deposition and fluence rate distribution across different lung diseases underscores the need for disease-specific treatment planning in PDT. In focal pneumonia, where collapsed regions absorb up to 3.5% of incident energy, high local absorption suggests that targeted PDT to consolidated zones could yield enhanced therapeutic selectivity and efficacy. In contrast, diffuse pneumonia, with more uniform deposition (~4.5% in collapsed areas), may require broader light coverage and adjusted fluence rates to ensure adequate treatment of the widespread involvement. COVID-19 cases present the most complex challenge: bilateral disease patterns, intermediate energy signatures in ground-glass opacities (~3%), and reduced fluence rate in dependent lobes collectively highlight the necessity of personalized strategies. Here, individualized illumination geometries, fluence rate adjustments, and possibly multiwavelength approaches could be explored to overcome the heterogeneous optical environment and improve treatment response.

From a clinical perspective, these findings emphasize that lung pathologies cannot be approached with a single irradiation protocol for PDT nor PBM. Instead, treatment strategies should be tailored to account for disease morphology, regional optical heterogeneity, and the specific optical fingerprints of diseased tissues. This has important implications not only for light delivery but also for photosensitizer dosing and treatment monitoring. For instance, the preserved vascular architecture, with minimal energy absorption, may offer routes for intravascular photosensitizer delivery without significant perturbation of light distribution. Meanwhile, the strong scattering observed in collapsed or infiltrated tissues highlights the need to carefully balance fluence rate to avoid overtreatment of dense regions while ensuring sufficient penetration into adjacent viable parenchyma.

On the other hand, in our simulations, all quantities were normalized to a source power of 1 W. To assess the clinical relevance of the calculated energy deposition, results must be scaled according to the irradiance at the skin surface (Is, W/cm²) and the exposure time (t). The therapeutic parameter of interest is the fluence, or energy density (ϕ₀, J/cm²), given by ϕ₀ = Is × t. Scaling the modeled outputs by the attenuation factors extracted from boxplots (Figs. 3 and 4) provides estimates of the energy density delivered to specific lung lobes. This step is essential to evaluate whether the modeled distributions meet practical dose requirements for PDT (25) or PBM (26).

Table 1 summarizes the therapeutic irradiance reaching the whole lung under Setup 4 at 808 nm, normalized to an irradiance at the skin of 100 mW/cm². Across all cases, mean lung irradiance values are on the order of 10⁻² mW/cm², with maxima between 1.6 and 3.9 mW/cm². Healthy lungs show higher transmission (mean: 0.0234 mW/cm²) compared to diseased states such as focal (0.0166 mW/cm²) or diffuse pneumonia (0.0120 mW/cm²). COVID-19 lungs exhibit similar reductions (mean: 0.0113 mW/cm²), consistent with the increased presence of collapsed or infiltrated regions. The large SD reflect spatial heterogeneity in energy deposition, with irradiance spanning from minimal values (~10⁻⁹ mW/cm²) up to several watts per square centimeter in localized regions.

Table 1.

Therapeutic irradiance at the whole lung from the setup 4 illumination at 808 nm (maximum, minimum, mean, and SD values)

Groups	Irradiance at the skin	max	min	mean	SD
Control volunteer	100 mW/cm2	3.94	1.49 × 10−9	0.0234	0.1065
Subject 1		1.66	1.40 × 10−9	0.0118	0.0644
Subject 2		2.38	1.47 × 10−9	0.0122	0.0644
Focal pneumonia		1.59	1.24 × 10−9	0.0166	0.0589
Diffuse pneumonia		3.64	1.31 × 10−9	0.0120	0.0833
COVID-19		2.63	2.36 × 10−9	0.0113	0.0659

These findings demonstrate that pathological lung changes significantly reduce light transmission, highlighting the need for accurate dosimetry in PDT and PBM. Using GPU-accelerated MC simulations with the Neu(t)ralMC framework, we modeled light propagation in CT-derived 3D lung geometries representing focal pneumonia, diffuse pneumonia, and COVID-19. The simulations revealed substantial differences in energy deposition and fluence-rate distributions driven by tissue consolidation, infiltration, and anatomical heterogeneity. These results underscore the importance of patient-specific light dosimetry and treatment planning to maximize therapeutic efficacy while minimizing damage to healthy tissue.

SI Appendix, Fig. S6 highlights the intersubject variability in absorbed energy distribution resulting from anatomical and morphological differences across the thoracic region. Despite identical illumination conditions, variations in lung geometry, chest wall thickness, and tissue composition led to markedly different internal light propagation patterns. This underscores the importance of patient-specific modeling for accurate light dose prediction and treatment planning in PDT. Altogether, these simulations integrate anatomical segmentation, optical property mapping, and MC modeling into a unified framework for personalized dosimetry and optimized light delivery strategies in pulmonary PDT.

An additional strategy that may further improve light delivery in pulmonary applications is optical tissue clearing. Previous studies have demonstrated that optical clearing approaches can transiently reduce tissue scattering and enhance light penetration in lung tissue through the administration of inhaled clearing agents or aerosols (62). Recent investigations have explored optical clearing for monitoring viral infections, multimodal assessment of lung injury, and three-dimensional visualization of intrapulmonary structures and cellular biodistribution (62, 63). Optical clearing has also been shown to modify the optical properties of lung tissue in a reversible manner and may therefore represent a potential approach for increasing light penetration during pulmonary phototherapy (64–66). Although optical clearing was not considered in the present simulations, future studies could incorporate experimentally measured changes in absorption and scattering coefficients associated with clearing protocols to evaluate their impact on intrapulmonary light distribution and treatment efficacy.

A translational workflow for patient-specific pulmonary phototherapy can be envisioned from this framework. First, high-resolution CT or MRI scans are used to segment healthy and diseased lung regions. Tissue-specific optical properties are then assigned using literature-based databases and pathology-dependent adjustments. GPU-accelerated MC simulations are subsequently performed to predict light propagation, fluence rate, and energy deposition throughout the lung. These outputs can be incorporated into treatment-planning algorithms to optimize illumination geometry and dosimetry prior to therapy. By combining voxelized anatomical models with advanced light-transport simulations, this approach supports personalized PDT and other photonic interventions. However, simulation accuracy depends on factors such as segmentation quality, voxel resolution, optical property assignment, and MC methodology. Future work should focus on experimental and clinical validation, multiwavelength modeling, and the integration of real-time dosimetry feedback to enable adaptive and individualized light-delivery strategies for pulmonary diseases.

Materials and Methods

Extended materials and methods are available in SI Appendix and include tissue volume definition, photon dosimetry metrics, light source positioning and region-of-interest selection, and optical probe calibration and cross-validation of the in-silico 3D model.

Data Collection and Ethical Considerations.

Four subjects from the Hospital das Clínicas, University of São Paulo School of Medicine, who had previously undergone chest CT examinations for clinical purposes, were included in this study. CT scans were acquired using a routine helical protocol with a peak tube voltage of 120 kVp and variable tube current (mAs) optimized by an automatic exposure control system. All images were reconstructed with a slice thickness of 1 mm. Intravenous contrast was administered only when clinically indicated.

The study cohort comprised one healthy control volunteer and three patients representing distinct pulmonary conditions: viral pneumonia caused by COVID-19, diffuse bacterial pneumonia, and predominantly focal bacterial pneumonia. The control subject presented no radiological evidence of pulmonary infection, lung disease, or abnormalities of the mediastinum or chest wall. The focal bacterial pneumonia case exhibited marked consolidation of the right lower lobe accompanied by additional infiltrates in the right middle and upper lobes. The diagnosis of COVID-19 pneumonia was confirmed by antigen testing together with characteristic radiological findings.

The study protocol was approved by the Research Project Review Ethics Committee (CAPPESQ) of the Hospital das Clínicas of FMUSP (approval No. CAAE: 71193623.1.0000.0068). Written informed consent was obtained from all participants prospectively enrolled in the study prior to inclusion. For deceased individuals whose CT data were retrospectively analyzed, the requirement for informed consent was waived by the Institutional Research Ethics Committee in accordance with applicable ethical regulations, as obtaining consent was not feasible.

The in vivo thoracic optical dosimetry measurements used for cross-validation of the computational model were performed in two healthy volunteers under the same study protocol approved. Written informed consent was obtained from both volunteers prior to participation and optical data collection.

All study procedures were conducted in accordance with institutional guidelines and the ethical principles of the Declaration of Helsinki, where applicable.

Image Segmentation.

In our CT scan segmentation workflow, we implemented a systematic approach to delineate various anatomical structures, with a combination of intensity thresholding, specialized segmentation tools, and advanced software extensions. The initial step involved extracting a 3D array of voxel intensity values, represented in Hounsfield Units (HU), from DICOM data for each tissue type. To enhance the visibility of lung structures, we used the window/level option in the Volumes module of 3D slicer (http://www.slicer.org) (33), using CT-Abdomen preset for tissue not related with lungs (Fig. 1 A, Left column), and CT-Lung preset for lung structures. For segmenting tissues with well-defined density ranges, such as skin, fat, muscle, and bones, we used intensity thresholding based on Hounsfield unit values. By setting specific thresholds for each tissue type in 3D Slicer’s Segment Editor, we generated masks that isolated these structures efficiently and with high accuracy. This approach allowed us to take advantage of the distinct radiodensities of each tissue, as documented in Table 2, achieving reliable segmentations for these anatomically simpler regions. For more complex anatomical structures such as the heart, liver, stomach, and kidneys, we utilized the TotalSegmentator extension (67) within 3D Slicer, designed specifically for CT and MR images. This tool enabled automated segmentation of major anatomical organs, improving the speed and consistency of results across scans. In cases where further refinement was necessary, a volume of interest (VOI) was delineated around the organ before applying intensity thresholds.

Table 2.

Threshold range of the HU scale for each biological tissue segmentation

Tissue type	Hounsfield unit intervals
Skin	[−500; −191]
Subcutaneous fat	[−190; −30]
Muscle	[−29; 149]
Bones	[150; 1,500]
Heart	[−29; 149] (TotalSegmentor)
Liver	[−29; 149] (TotalSegmentor)
Stomach	[−29; 149] (TotalSegmentor)
Kidney	[−29; 149] (TotalSegmentor)
Lungs	[−1,050; −750] (Lung CT Segmenter and Analyzer)
Collapsed	[−400; 0.0] (Lung CT Segmenter and Analyzer)
Infiltrate	[−750; −400] (Lung CT Segmenter and Analyzer)
Blood vessels	[0.0; 1,000] (Lung CT Segmenter and Analyzer)

Lung segmentation required a multistep, nuanced approach due to the complexity and internal variability within lung tissues. Initially, the Airway Segmentation extension (68) in 3D Slicer (https://github.com/Slicer/SlicerAirwaySegmentation/) was used to isolate the trachea by placing a markup point in the trachea and segmenting based on input volume. Following this, we used the Lung CT Segmenter extension (https://github.com/rbumm/SlicerLungCTAnalyzer/), which enabled detailed segmentation of lung lobes and airways. Using 3D Slicer’s AI model options, particularly the lungmask U-net function (69) that leverages a U-net model, we specified the input segmented volume of the lungs to automate segmentation of these areas. Once primary lung masks were generated, we employed the Lung CT Analyzer extension to segment critical internal structures, including blood vessels, inflated regions, infiltrated tissues, and collapsed regions. This multilayered approach allowed us to capture essential structural variations within the lung, supporting our pipeline’s goal of creating highly accurate tissue models for MC simulation and other analytical processes.

Optical Properties Library.

Light–tissue interaction is characterized by the absorption coefficient (μ_a), scattering coefficient (μ_s), scattering anisotropy (g), and refractive index. In this study, wavelength-dependent optical properties assigned to each segmented tissue were derived from comprehensive literature reviews (14, 29, 30, 32, 70), and were assumed to be spatially homogeneous within each tissue class. These values do not account for patient-specific variations in chromophore composition, blood content, oxygenation status, or disease-related changes, and therefore should be considered as approximations for dosimetry estimation.

In tissue, absorption in the visible range is predominantly influenced by oxy- and deoxy-hemoglobin (HGb), while water absorption becomes significant in the near-infrared range. Additional absorbers include beta carotene (in fat), melanin (in hair and skin), bile (in the liver), and collagen. The total absorption of a tissue is calculated from the linear combination of the various absorption spectra, scaled by the concentration of each chromophore. Any tissue can be characterized by the following equation:

$$\begin{array}{c}{\mu }_a\left(\lambda \right)=B·S·{\mu }_{a, oxy}\left(\lambda \right)+B·\left(1-S\right)·{\mu }_{a, deoxy}\left(\lambda \right)+W·{\mu }_{a, water}\left(\lambda \right)\hfill \\ +F·{\mu }_{a, fat}\left(\lambda \right)+M·{\mu }_{a, melanosome}\left(\lambda \right),\end{array}$$ [1]

where S is the HGb oxygen saturation, B the average blood volume fraction, W the water content, F the fat content, M the melanosome volume fraction. The values for these parameters for different tissue types are provided in Table 3.

Table 3.

Biological tissue parameters governing the optical absorption for different tissue types

Tissue type	B%	S%	W%	M%	F%	Reference
Skin	0.20	67	65	2.50	0	From (29) we used: Choudhury et al. (2010)
Subcutaneous fat	0.80	21	35	0	65	From (29) we used: Weaver et al. (1989)
Muscle	0.80	21	82	0	18	From (29) we used: Weaver et al. (1989)
Bones	1.20	72	31	0	80	From (29) we used: Weaver et al. (1989)
Heart	1.00	90	50	0	20	(29)
Liver	3.20	8.0	76	0	19	(29)
Kidney	1.00	90	50	0	20	(29)
Esophagus/Stomach	3.60	90	70	0	5	(29)
Lungs	17.0	67	58	0	0	From (29) we used: Weaver et al. (1989)
Collapsed*						(70)
Infiltrate	Water absorption spectrum					(29)
Blood vessels	100	75	95	0	0	(29)

^*The optical properties of collapsed lung tissue were approximated using values reported for articular cartilage. This assumption is based on the structural similarities between collapsed alveoli and dense, water-rich connective tissue: both exhibits reduced air fraction, high extracellular matrix content, and relatively homogeneous scattering behavior. In the absence of direct measurements for collapsed lung tissue, cartilage provides a conservative estimate for absorption and scattering coefficients.

Unlike absorption, there is no well-defined analytical relationship to describe the spectral dependence of volume scattering. However, based on experimental observations and simplifying analysis of Mie Scattering Theory, there is a consensus that the reduced scattering coefficient can be modeled using a power law relationship:

$${{\mu }^{\prime }}_s(\lambda )={{\mu }^{\prime }}_s(500 \mathrm{nm})\left[f_{\mathit{Ray}}{\left(\frac{\lambda }{500 \mathrm{nm}}\right)}^{-4}+\left(1-f_{\mathit{Ray}}\right){\left(\frac{\lambda }{500 \mathrm{nm}}\right)}^{-b_{\mathit{Mie}}}\right],$$ [2]

where f_Ray is the fraction of scattering attributed to the contribution of Rayleigh scattering, b_Mie is the scattering power for Mie scattering determining how the scattering coefficient changes as a function of wavelength. The parameters for the reduced scattering coefficient for different tissue types are provided in Table 4.

Table 4.

Biological tissue parameters governing the reduced scattering coefficient for different tissue types

Tissue type	${{\mu }^{\prime }}_s\left(500 \text{nm}\right)$	$f_{Ray}$	$b_{Mie}$	g	Reference
Skin	42.2	0.62	1.00	0.90	From (29) we used: Choudhury et al. (2010)
Subcutaneous fat	14.1	0.17	0.53	0.90	From (29) we used: Weaver et al. (1989)
Muscle	13.0	0.13	0.93	0.90	From (29) we used: Weaver et al. (1989)
Bones	38.4	–	1.47	0.90	From (29) we used: Weaver et al. (1989)
Heart	8.30	–	1.26	0.90	(29)
Liver	16.5	–	1.64	0.90	From (29) we used: Yi and Backman (2012)
Kidney	35.1	–	1.51	0.90	From (29) we used: Alexandrakis et al. (2005)
Esophagus/stomach	19.1	0.02	1.09	0.95	(29)
Lungs	8.10	–	0.98	0.90	From (29) we used: Weaver et al. (1989)
Collapsed				0.90	(70)
Infiltrate	Water absorption spectrum			0.82	(29)
Blood vessels	10.0	0.0	0.90	0.90	(29)

This library of optical properties was essential for modeling light–tissue interactions accurately in our MC simulations, enabling precise dosimetry calculations necessary for effective and safe PDT and PBM protocols.

Modeling of Photon Transport in Turbid Media.

The effectiveness of light-based methods in turbid media such as biological tissues is strongly limited by multiple scattering of light, which degrades imaging quality and therapeutic performance (71). Designing optical systems for noninvasive in-vivo treatment therefore requires a deep understanding of light–tissue interactions through in-silico modeling, incorporating parameters such as wavelength, coherence, polarization, beam profile, and geometry of detector. Due to analytical solutions being rarely feasible in heterogeneous biological media, stochastic approaches, particularly MC simulations, have become the “gold standard” across biomedical optics and related fields (72–74), evolving significantly alongside advances in computational hardware.

In this work we employ an open-source, power-efficient NMC framework developed in-house that achieves significant energy savings and speed compared with traditional implementations (43, 74). Optimized for modern ARM-based architectures, our model leverages unified memory, enabling large-scale photon transport simulations on detailed anatomical models of thoracic cavity without performance compromises. NMC supports extensive customization and advanced optical phenomena including polarization and orbital angular momentum (OAM) (75), and has been validated against established biomedical solvers, and prior established experimental datasets (76).

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Study data are included in the article and/or SI Appendix. MATLAB scripts, optical property library, simulation parameters, representative Monte Carlo simulation files, processed datasets underlying the figures, and documentation supporting this study have been deposited in a publicly accessible GitHub repository (https://github.com/JohanDiazTovar/Monte-Carlo-Modeling-of-Light-Propagation-in-Healthy-and-Diseased-Human-Lungs.git) (77). Original clinical CT images are not publicly available because of institutional ethics and patient privacy restrictions.