Molecular modelling-driven formulation design for targeted pulmonary drug delivery.

Interstitial Lung Disease (ILD) is a collection of progressive pulmonary conditions that impact on the pulmonary tissue structure and progressively diminish respiratory function. Early identification of ILD is still a difficult task since radiological differences are delicate and may be confused with other respiratory diseases. In the recent past, artificial intelligence has been developed to analyze medical images automatically and this has offered a new chance of enhancing diagnostic accuracy. The proposed study will develop a hybrid framework of deep learning based on Vision Transformer (ViT) and Convolutional Neural Network (CNN) structures to performearly detecting of ILD using chest CTscansand X-rays. The CNN element is concerned with the extractionoffine-scale localspacefeatures,including the texture anomalies and fibrotic structures, the Vision Transformer is concerned with the global contextual relationship between lung regions and other regions through the self-attention mechanisms. The representations extracted are combined to come up with an integrated prediction model that can differentiate between lungs with ILD and normal ones. A web-based clinicalsupport system iscreatedto facilitatethe real-time predictionbygivingmedicalpractitionersanopportunity to post-imaging data and receive automated diagnostic information. As shown in the experiments, the hybrid architecture suggested is better at classification than the singlemodel solutions, especially on the detection of early-stage abnormalities.

Electrical impedance tomography (EIT) has been proposed as a bedside method of measuring pulmonary blood patterns in neonates. However, EIT's use has been limited by technical issues including cardiac-motion artefact and ventilation masking the heartbeat-related signals (HRS). This study aims to determine EIT-derived pulmonary blood volume patterns in two distinct biological models. The HRS were extracted from EIT recordings during Study A: 10 ml saline administered into the right atrium of apnoeic lambs (n = 6) with cardiac output, and during asystole. Study B: A sustained inflation (aeration without tidal ventilation) during the respiratory transition at birth in preterm lambs (n = 12). Study A: There was a significant fall in HRS-impedance within lung regions upon administration of saline that was independent of cardiac output (p < 0.0001, Two-way repeated-measure ANOVA), which was not influenced by cardiac motion (p = 0.16), suggesting EIT can detect changes to pulmonary arterial bed blood volume. Study B: The amplitude of the HRS decreased linearly during aeration (R2 = 0.062; linear regression) as a net result of decreased diastolic and increased systolic HRS. EIT demonstrated expected pulmonary blood volume changes in the lung independent of cardiac motion, highlighting EIT's potential to measure ventilation and pulmonary circulation mismatch in neonates. EIT may represent a future potential solution to the lack of non-invasive and radiation-free measurements of ventilation and pulmonary perfusion in neonates. Our study showed that EIT can detect changes in thoracic impedance that are independent of ventilation and heart motion within the chest and can delineate pulmonary vascular changes that occur at birth. This suggests EIT could be used to detect blood volume changes in the lungs of neonates in the immediate postnatal period.

Sickle cell disease (SCD) causes pulmonary parenchymal and vascular complications with a major impact on mortality. Quantitative computed tomography (CT) assessment of pulmonary vascular "pruning" (rarefaction of small-caliber pulmonary vessels) may provide a non-invasive tool to better understand these complications. To evaluate vascular pruning in SCD and its association with clinical, functional respiratory, and imaging parameters. Non-contrast chest CT scans from 73 adult patients with SCD followed at Avicenne Hospital (Paris, France) were analyzed. Pulmonary vascular volumes were quantified using the blood volume in vessels with a cross-sectional area < 5 mm² (BV5) and the ratio of BV5 to total pulmonary blood volume (TBV), measured globally and in peripheral lung regions. These biomarkers were compared with pulmonary function tests (PFTs), CT parenchymal abnormalities, and clinical history of vascular disease. In 73 patients (mean age 33 ± 14 years; 38 women), linear opacities (78%), reticulations (53.4%), and proximal vascular abnormalities (pulmonary trunk enlargement and segmental artery dilation: 28.8% and 19.2%, respectively) were significantly associated with reduced BV5 and BV5/TBV, indicating increased pruning. Similar associations were observed in patients with a hemoglobin-corrected diffusing capacity for carbon monoxide (DLCOc) < 80%, history of pulmonary embolism (PE), and pulmonary hypertension (PH). No significant relationship was found with acute vaso-occlusive episodes. Peripheral lung analysis showed trends consistent with global measurements, with some differences in statistical significance. Pulmonary vascular pruning is observed in SCD in association with parenchymal and vascular abnormalities and may serve as an early imaging biomarker of chronic vasculopathy. Question Can automated quantitative analysis of pulmonary vessels on non-contrast chest CT reveal vascular pruning and provide imaging biomarkers of pulmonary vasculopathy in sickle cell disease? Findings CT-derived vascular pruning markers (BV5 and BV5/TBV) were associated with reduced DLCOc, history of pulmonary hypertension or embolism, and parenchymal and vascular abnormalities in SCD. Clinical relevance Quantification of vascular pruning on routine non-contrast chest CT may provide a practical, non-invasive biomarker for detecting pulmonary vasculopathy and identifying sickle cell disease patients at risk for pulmonary hypertension.

Sepsis-induced acute lung injury (ALI) is characterized by severe oxidative stress and senescence of alveolar type II epithelial cells (AECⅡ), yet effective therapies capable of reversing these pathological processes remain lacking. We developed a multifunctional ICAM1-targeted MnO2@ZIF8 nanoplatform encapsulating Tanshinone IIA (TSA) that integrates selective targeting, redox regulation, and senescence reprogramming. The nanoplatform preferentially accumulated in inflamed lung regions, efficiently scavenged reactive oxygen species (ROS), activated the Nrf2-mediated antioxidant pathway, and suppressed the IL-33/ST2 inflammatory axis. Single-cell RNA sequencing revealed that treatment reshapes AECⅡ differentiation trajectories, suppresses senescence-related transcriptional signatures, and enhances alveolar regenerative potential by restoring stem-like phenotypes. Proteomic and metabolomic analyses further confirmed improved mitochondrial metabolism, SASP modulation, and inflammatory resolution. Functionally, Anti-ICAM1-MnO2@ZIF8@TSA restored alveolar structure, boosted oxygenation, and markedly increased survival in CLP-induced septic mice. Collectively, these findings present a nanobiotechnological strategy that rejuvenates senescent epithelial cells and provides a mechanistically guided therapeutic approach for the treatment of sepsis-associated pulmonary failure.

The administration of contrast material increases the density of lunge parenchyma on computed tomography. Lung pathologies which are characterized by an only slight alteration of lung attenuation (e.g. ground glass opacities or emphysema) should therefore be evaluated on non-enhanced scans in order to avoid misinterpretation. However, contrast administration is very helpful in some clinical scenarios like suspected pulmonary embolism or malignancy. This study aimed to quantify the amount of increase of lung parenchymal density after intravenous contrast administration compared to non-enhanced images and whether this increase exhibits regional variation similar to or different from the attenuation on non-enhanced scans. This retrospective, IRB-approved, bi-center study included patients who underwent both contrast-enhanced dual-energy and non-enhanced chest CT scans within a year between 04/2018 and 12/2022. Scans were co-registered and semi-manually segmented into the whole lung and isovolumetric segmentations of the ventral/dorsal halves and upper/middle/lower thirds. Mean lung density for each region was calculated from contrast-enhanced (CE), virtual non-contrast (VNC), and true non-contrast (TNC) scans. Bland-Altman analyses with non-parametric limits of agreement assessed the mean difference in attenuation values, with differences between lung regions tested by the Mann-Whitney U test. Subgroup differences were also analyzed by Mann-Whitney U tests. Correlation analyses were performed to investigate the correlation between the increase in density from contrast media and the non-enhanced density. Fourty eight patients (26 females, median age: 63y) fulfilled the inclusion criteria. The majority suffered from emphysema and/or overinflation. The mean increase in lung density between TNC and CE scans was 11.65 HU [95% Confidence Interval: 6.10, 17.19] for the whole lung. Amount of increase ranged from 7.79 HU [0.65, 14.94] to 16.35 HU [11.22, 21.53] for different lung regions. Comparing VNC with CE scans showed an average increase in lung density of 11.21 HU [8.63, 13.79]. Amount of increase ranged from 2.02 HU [-2.06, 6.10] to 16.58 HU [13.47, 19.70] for different lung regions. A significant difference between contrast enhancement based on TNC versus VNC images was found for the upper third only. For the whole lung and all other lung regions no differences in density increase were seen. Regression analyses and Spearman's Rho showed that increase in density tended to be more pronounced in lung regions with higher baseline density in non-enhanced scans. The mean increase in density of lung parenchyma after contrast administration was approx. 11 HU, for both TNC and VNC baseline images. There were substantial variations across individuals and lung regions with a tendency of higher increases in areas with higher baseline density on non-enhanced images. Knowledge of these phenomena avoid misinterpretation of contrast enhanced scans.

Lung cancer remains one of the leading causes of cancer-related deaths worldwide, highlighting the urgent need for accurate and interpretable diagnostic tools. While deep learning (DL) models have achieved strong results in medical image classification, their opaque decision-making process remains a barrier to clinical adoption. This study proposes an adaptive superpixel perturbation-based local interpretable model-agnostic explanations (ASP-LIME), a novel explanation framework designed to generate faithful and localized interpretations of DL predictions, providing insights into the model's decision-making process. The proposed approach improves upon the original local interpretable model-agnostic explanations method by introducing adaptive superpixel segmentation, stratified perturbation strategies, lung region masking, and post-processing enhancements tailored for medical imaging. The proposed framework is applied to a lung cancer classification task using a custom-designed convolutional neural network, MedDeepNet, as the predictive model. Experimental results on a publicly available lung image dataset demonstrate that MedDeepNet achieves 99.84% accuracy, 99.66% recall, 99.82% precision, 99.74% specificity, and a 99.74% F1-score. ASP-LIME produces high-fidelity explanations with strong localization to pathological regions, achieving scores of 0.0300 for deletion, 0.9622 for insertion, and 0.9661 for Area Between Perturbation Curves (ABPC), surpassing typical benchmarks for interpretability methods. The findings demonstrate that the proposed framework offers consistent and interpretable explanations that enhance understanding of model decisions in medical imaging applications.

Mitochondrial damage is a conserved feature of coronavirus infection, occurring with human (SARS-CoV-2, HCoV-OC43) and murine (MHV-1) coronaviruses. Coronaviruses damage mitochondria in airway epithelial cells (AEC), pulmonary artery smooth muscle cells (PASMC), pulmonary artery endothelial cells, immune cells and cardiomyocytes by causing rapid transcriptomic changes in nuclear-encoded genes regulating mitochondria and by viral proteins interacting with host mitochondrial proteins. Coronavirus infection causes mitochondrial depolarization, mitochondrial transition pore (MTP) opening, inhibition of the electron transport chain (ETC) and ATP synthetic apparatus, increased mitochondrial fission, apoptosis, and impaired mitochondrial oxygen sensing. Within hours of infection, SARS-CoV-2 induces transcriptional reprogramming of genes relevant to the mitochondrial matrix in AECs, downregulating mRNA encoding ETC complex I components and the ATP synthesis complex. These bioenergetic consequences of SARS-CoV-2mitochondriopathy may contribute to long COVID. Infection also upregulates dynamin-related protein 1 (DRP1), activating mitochondrial fission while promoting apoptosis by activating apoptosis inducing factor (AIF) and caspase 7. Even without infection, transfection with specific coronaviral proteins opens the MTP and depolarizes the mitochondria, or activates DRP1 and AIF, promoting AEC damage or apoptosis, thereby contributing to diffuse alveolar damage. In human PASMCs, coronaviral M and Nsp9 proteins suppress hypoxic pulmonary vasoconstriction (HPV), a homeostatic mechanism in PASMCs that uses a mitochondrial oxygen sensor to redistribute blood flow to well-ventilated lung regions during pneumonia. Impairment of HPV, seen as intrapulmonary shunting, contributes to the profound hypoxaemia in COVID-19 pneumonia. Coronavirus-induced mitochondriopathy may have therapeutic relevance as blocking AIF-induced apoptosis or enhancing HPV appears beneficial in a MHV-1model of COVID-19 pneumonia.

Lung cancer segmentation Using the Att-U-Net Model on PET-CT Images.

Subclavian venous access may result in pleural injury. During positive-pressure ventilation (PPV), increased intrathoracic pressure compresses peripheral lung tissue, promoting closure of terminal airways. As a result, air leakage from injured lung regions may remain confined and not communicate with the pleural space, rendering intraprocedural and early post-procedure imaging falsely negative. We report a 67-year-old woman in whom pneumothorax became radiographically apparent only after extubation. The transition to spontaneous (negative-pressure) breathing reduced surrounding pressure, allowing reopening of peripheral airways and facilitating the movement of air into the pleural space. This case provides a mechanistic explanation for delayed radiographic detection of pneumothorax under PPV and supports the use of planned post-extubation imaging when pleural injury is suspected.

In India, Lung diseases are a major health crisis, with Pulmonary Fibrosis and pneumonia being the most prevalent, making India a global hotspot for respiratory issues. Recent global health data indicate that respiratory infections like pneumonia remain a leading cause of mortality, accounting for over 2.4 million deaths annually, while interstitial lung diseases such as pulmonary fibrosis show a rising prevalence of approximately 30 to 70 cases per 100,000 people. The need for automated diagnostic systems is paramount in emergency departments and rural clinics where 24/7 access to expert thoracic radiologists is often unavailable. Such applications provide a vital second-opinion tool that can triage urgent cases of pneumonia or track the progression of chronic conditions like pulmonary fibrosis in resource-limited settings. While classifying C-XR images as Normal, Pneumonia and Pulmonary Fibrosis, Traditional manual interpretation of chest X-rays is frequently hindered by high inter-observer variability and a significant risk of human fatigue during high-volume shifts. Furthermore, subtle earlystage lesions or complex fibrotic patterns can be easily overlooked by non-specialists, leading to delayed treatment or misdiagnosis. The proposed methodology utilizes the C-XR datasets, which provide a robust collection of labelled images for Normal, Pneumonia, and Pulmonary Fibrosis classes. The proposed system implements a Vision Transformer (ViT) for feature extraction, which discards traditional convolutional layers in favour of a self-attention mechanism. This approach breaks the CXR image into fixed-size patches and uses an encoder to capture global spatial dependencies and longrange relationships between distant lung regions. While existing benchmarks often rely on Multi-Layer Perceptron (MLP), Random Forest Classifier (RFC), and Extreme Gradient Boosting (XGB) had lower performance, so this research proposes the integration of a Light Gradient Boosting Machine (LGBM). It is selected for its leaf-wise tree growth strategy and histogram-based binning, which significantly reduces training time and memory consumption while maintaining superior accuracy on the highdimensional feature vectors produced by the Transformer.

Supraglottic jet ventilation (SJV) is used during flexible bronchoscopy to improve oxygenation, but its impact on overall lung ventilation is unclear. Thoracic electrical impedance tomography (EIT) offers real-time monitoring of respiratory function. This prospective pilot study enrolled patients undergoing flexible bronchoscopy. End-expiratory lung impedance (EELI) changes from baseline (ΔEELI) were measured at baseline (T0), oxygenation with Wei nasal jet (WNJ) (T1), SJV (T2), during bronchoscopy (T3), and post-bronchoscopy (T4). The primary outcome was ΔEELI at T2 vs. T3. Secondary outcomes included ΔEELI at other time points, ventilation distribution (ventral vs. total lung, ROIventral), hemodynamics, oxygen saturation (SpO2) at each point, and adverse events. Among 27 patients, 5 experienced transient desaturation (no laryngeal mask ventilation needed). ΔEELI and SpO2 significantly decreased from T2-T3 (ΔEELI: -2.65, 95% CI -4.27 to -1.02, p < 0.001; SpO2: -2.5%, 95% CI -4.7 to -0.3%, p < 0.001). Conversely, ΔEELI and SpO2 significantly increased from T1-T2 (ΔEELI: 1.84, 95% CI 1.02-2.66, p < 0.001; SpO2: 3.3%, 95% CI 1.3-5.2%, p < 0.001), with ROIventral rising from 43.7 to 60.4% (16.6% increase, 95% CI 4.4-28.9%, p < 0.001). While SJV did not prevent the negative effects of bronchoscopy on lung ventilation and oxygenation, it maintained adequate SpO2. Pre-bronchoscopy SJV improved these measures, but shifted ventilation towards ventral lung regions compared to WNJoxygenation. Chinese Clinical Trial Registry (URL: Chictr.org.cn), Identifier: ChiCTR2100050285. Date of registration: Aug 25th, 2021.

ABSTRACTCardiac arrest due to severe chest trauma with thoracic aortic injury presents a significant challenge in emergency medicine. Extracorporeal cardiopulmonary resuscitation (ECPR) offers a potentially lifesaving intervention for such cases. We report a case of a 22‐year‐old male who sustained an open chest trauma with thoracic aortic injury from a stab wound, resulting in cardiac arrest. The patient was treated with ECPR combined with emergent surgical repair. Electrical impedance tomography (EIT) monitoring provided real‐time functional assessment of regional lung mechanics during the 8‐day ECMO support period. Serial chest radiographs documented improvement following bedside hematoma evacuation, with improvement in PaO2/FiO2 ratio from 118 to 442.5. The patient achieved successful ECMO weaning; however, due to prolonged cardiac arrest time exceeding 30 min, neurological recovery was not achieved (GCS remained 3 T), and the family withdrew care after 2 months. Despite the unfavorable neurological outcome, this case demonstrates the technical feasibility of ECPR in severe penetrating chest trauma and provides valuable experience for managing similar cases with potentially shorter arrest times.

Sex-dependent differences in regional lung mechanics; A retrospective observational study.

COVID-19 still poses a global public health challenge, exerting pressure on radiology services. Chest X-ray (CXR) imaging is widely used for respiratory assessment due to its accessibility and cost-effectiveness. However, its interpretation is often challenging because of subtle radiographic features and inter-observer variability. Although recent deep learning (DL) approaches have shown strong performance in automated CXR classification, their black-box nature limits interpretability. This study proposes an explainable deep learning framework for COVID-19 detection from chest X-ray images. The framework incorporates anatomically guided preprocessing, including lung-region isolation, contrast-limited adaptive histogram equalization (CLAHE), bone suppression, and feature enhancement. A novel four-channel input representation was constructed by combining lung-isolated soft-tissue images with frequency-domain opacity maps, vessel enhancement maps, and texture-based features. Classification was performed using a modified Xception-based convolutional neural network, while Gradient-weighted Class Activation Mapping (Grad-CAM) was employed to provide visual explanations and enhance interpretability. The framework was evaluated on the publicly available COVID-19 Radiography Database, achieving an accuracy of 95.3%, an AUC of 0.983, and a Matthews Correlation Coefficient of approximately 0.83. Threshold optimisation improved sensitivity, reducing missed COVID-19 cases while maintaining high overall performance. Explainability analysis showed that model attention was primarily focused on clinically relevant lung regions.

Wood smoke (WS)-derived PM2.5 is a major environmental risk factor for emphysema, but the role of macrophage-derived exosomes remains unclear. Here, we show that chronic WS exposure in rats induces emphysematous lesions accompanied by markedly increased exosome release, reflected by elevated CD63 expression in alveoli and increased exosome particle number in bronchoalveolar lavage fluid. Pharmacological inhibition of exosome secretion with GW4869 significantly attenuated alveolar destruction. Moreover, intranasal administration of exosomes from PM2.5-exposed macrophages was sufficient to recapitulate emphysematous pathology in vivo. Mechanistically, PM2.5 exposure upregulated the membrane-trafficking protein SNAP23 in macrophages, enhancing exosome secretion and increasing total abundance of ADAM10 within exosomes. Co-localization of CD63 and ADAM10 was observed in alveolar regions of WS-exposed rats and COPD patient lungs. Silencing ADAM10 suppressed caspase-3 activation and reduced epithelial apoptosis, supporting a role for exosomal ADAM10 in driving caspase-3-dependent alveolar epithelial injury. By defining an exosome-mediated macrophage-epithelium pathway that contributes to PM2.5-induced emphysema, this study clarifies how combustion-derived PM2.5 damages lung structure and offers mechanistic insight to reducing the health burden of biomass-related air pollution.

Objective.This study aims to develop a functional-based multi-omics model for early prediction of radiation pneumonitis (RP) by extracting radiomic and dosiomic features from functionally defined lung regions, using generated perfusion (Q) and ventilation (V) from pre-radiotherapy planning computed tomography (CT).Approach.We retrospectively analyzed data from 121 patients with locally advanced non-small cell lung cancer treated with curative-intent intensity-modulated radiotherapy between 2015 and 2019, including pre-treatment CT and dose maps. Q and V maps were generated from CT with deep learning-based and supervoxel-based approaches, respectively. Regions of interest (ROIs) combined the planning target volume with each of three functional lung regions-high functional lung (HFL), low functional lung, and whole lung (WL)-defined by thresholds on Q and V maps. Radiomic and dosiomic features were extracted from CT and dose distributions within each ROI. For each ROI, three methods-radiomics (R), dosiomics (D), and dual-omics (RD)-were constructed. 13 machine learning algorithms were trained and evaluated using 10-fold cross-validation, and model performance was assessed by the average area under the receiver operating characteristic curve (AUC), accuracy, precision, recall, andF1 score. RP was defined as CTCAE grade ⩾2.Main results.Of the 35 selected features, 20 were from HFL. In dual-omics models, using HFL features improved predictive performance for RP (AUC 0.879 ± 0.105) compared to WL (AUC 0.778 ± 0.100). In HFL, the RD method outperformed both R (AUC 0.786 ± 0.076) and D (AUC 0.791 ± 0.107) methods. Decision curve analysis showed the dual-omics model based on HFL provided the highest net benefit across threshold probabilities.Significance.This study is the first to systematically demonstrate that features extracted from CT-derived HFL capture important functional differences and provide strong predictive value for RP. Compared to conventional methods, integrating radiomics, dosiomics, and CT-based functional information further improves predictive performance.

Pulmonary ventilation imaging enables functional avoidance radiotherapy treatment plans by quantifying regional lung function. However, current clinical standards, such as 99𝑚Tc-based single-photon emission computed tomography (SPECT), rely on radioactive tracers, which can introduce imaging deposition artifacts. CT ventilation imaging (CTVI) methods based on both physical models and deep learning approaches currently require multiple CT images as input, such as the inhale/exhale phases of a 4DCT. While the theoretical foundation of physics-based CTVI is built on multi-phase information, the feasibility of single-phase deep learning CTV models has not been determined. While deep learning methods have predicted SPECT ventilation from multi-phase 4DCT, the benefit of including more than one respiratory phase remains unclear. Predicting ventilation using only single-phase CTs reduces computational expense, potentially simplifies the image acquisition process, and avoids artifacts introduced by image registration, thereby making deep learning-based CTV approaches more feasible for clinical applications outside of radiotherapy. This study (1) develops a deep learning model to predict SPECT ventilation using only the inhale phase of non-contrast 4DCT and (2) evaluates the impact of adding the exhale phase. We developed a SwinUNETR-based architecture using the maximum inhale 4DCT phase to predict pulmonary ventilation. A total of 44 cases with paired inhale CT and SPECT scans were used in the training. To assess multi-phase benefits, we compared: (1) InhaleCT-Swin Model-trained on inhale CT only; (2) ExhaleCT-Swin Model-trained on exhale CT only; (3) Hybrid Models IECT-Swin-FTD, IECT-Swin-FTDE, IECT-Swin-FTDES, fine-tuned on inhale/exhale CT pairs (IECT) with varying network components updated. A standard U-Net was also trained on inhale CT (InhaleCT-UNet), exhale CT (ExhaleCT-UNet), and IECT (IECT-UNet) for cross-architecture evaluation. The SwinUNETR-based Hybrid Model, IECT-Swin-FTD, achieved mean voxel-wise Spearman correlation of 0.762±0.035, outperforming the current state-of-the-art methods. Our transformer-based model trained on inhale CT slightly outperformed exhale CT with no significant differences ( ). U-Net achieved lower overall accuracy, though its highest performance occurred with IECT. No significant difference was found between InhaleCT-Swin Model and the best-performing hybrid UNet Model, IECT-UNet ( ). A transformer-based model with its decoder fine-tuned on IECT (IECT-Swin-FTD) achieved state-of-the-art accuracy for SPECT ventilation prediction. Moreover, our InhaleCT-Swin Model achieved comparable results with widely used UNet-based models that require multi-phase CT, showing that single CT may be sufficient for accurate ventilation prediction and may improve clinical workflow by reducing acquisition requirements and registration-related artifacts.

Mechanical ventilation (MV) plays a vital role in intensive care, ensuring sufficient gas exchange in acute respiratory distress syndrome (ARDS) patients. However, ventilator-induced lung injury (VILI) remains a frequent complication associated with MV, arising due to local lung tissue hyperinflation (HI) and cyclic alveolar recruitment/derecruitment (R/D). Determining optimal ventilator settings is a clinical challenge, since the full spectrum of local lung mechanics in a heterogenous lung cannot be assessed with overall mechanical measurements, nor with routine imaging modalities. Computational modelling offers a promising approach for personalizing mechanical ventilation settings, by predicting the local lung mechanical behavior. We propose an in silico model of the respiratory system of a mechanically ventilated ARDS patient, which integrates local patient-specific lung characteristics. These include both structural (airway tree and lung morphology) and functional (regional lung elastance and R/D dynamics) information, inferred from computed tomography (CT) data obtained at two different respiratory pressure instances. Our proof-of-principle simulations indicate that the model plausibly estimates the global respiratory pressure-volume curve, as well as regional lung biomechanical behavior, under positive pressure ventilation. Further, we show that this model can be used to simulate the effect of changes in ventilator settings such as positive end-expiratory pressure (PEEP), or to simulate an impaired lung with worsening biomechanics. This model thereby provides a mechanistic foundation to eventually support clinicians in delivering more precise, patient-specific therapies, by offering a supplementary tool for optimizing ventilator settings.

Background: Chest X-ray (CXR) image classification remains a critical tool in COVID-19 and pneumonia diagnosis. While segmentation of lung fields is commonly assumed to improve deep learning classification performance, recent evidence suggests that segmentation may also remove clinically relevant contextual cues. Attention mechanisms, particularly those that combine spatial and channel information, have shown potential to enhance model focus and generalizability. Objective: This study investigates the effectiveness of the proposed Self-Adaptive Convolutional Block Attention Module (SA-CBAM) in improving CXR image classification performance when tested on both segmented and unsegmented images. In addition, the benefit of hybridization of Convolutional Neural Network (CNN) with a Long Short-Term Memory (LSTM) network is evaluated. Methods: A U-Net segmentation model was used to extract lung regions from the COVID-QU-Ex dataset. Multiple model variants were implemented and compared, including a baseline CNN, CNN models integrated with several channel–spatial attention mechanisms, and the proposed SA-CBAM. In addition, hybrid architectures combining CNN with LSTM networks were evaluated. All models were assessed on both segmented and unsegmented CXR pipelines using accuracy, recall, specificity, F1-score, and Matthews Correlation Coefficient (MCC) as primary performance metrics. Results: Models trained on unsegmented CXR consistently outperformed those trained on segmented images. The proposed CNN-SA-CBAM improved baseline CNN performance from 88.42% to 90.08% accuracy and from 82.82% to 85.36% MCC when trained on unsegmented data. Further hybridization with an LSTM network produced the highest performance, where the CNN-SA-CBAM-LSTM achieved 99.90% accuracy and 99.85% MCC on unsegmented CXR. Despite segmentation producing visually clearer attention maps, classification performance was lower, suggesting potential loss of subtle contextual cues outside the segmented lung boundaries. Conclusion: The findings demonstrate that SA-CBAM significantly enhances CXR classification performance, particularly when applied to unsegmented images and further combined with LSTM. This study challenges the common assumption that segmentation always improves classification accuracy and highlights the importance of preserving contextual information. Future work will focus on adaptive or soft segmentation strategies that retain peripheral cues while preserving anatomical interpretability.