A medical infographic and diagnostic algorithm in Portuguese for risk stratification of acute pulmonary embolism (APE). At the center is a diagnostic image: a transverse/axial CT pulmonary angiogram (CTPA) showing the mediastinum and major hilar vessels, including the pulmonary artery and heart chambers. The algorithm branches into three risk categories: 'Baixo Risco' (Low Risk), defined by a PESI score ≤ II and absence of right ventricular (RV) dilation; 'Alto Risco' (High Risk), characterized by circulatory shock (SBP < 90 mmHg) or cardiorespiratory arrest; and 'Risco intermediário' (Intermediate Risk). The intermediate section details an association of biomarkers across three scoring systems: the BOVA score (incorporating SBP, troponin, RV dysfunction, and heart rate), the TELOS score (lactate, troponin, and RV dysfunction), and the SHIELD score (shock index, lactate, hypoxemia, and RV dysfunction). Arrows indicate the clinical flow based on positive or negative scores, integrating clinical vitals, imaging findings of RV strain, and laboratory markers for triage and management.

This composite educational graphic combines diagnostic imaging and a clinical decision-making algorithm for the risk stratification of acute pulmonary embolism (PE). At the center is an axial view of a Contrast-Enhanced Computed Tomography (CT) scan of the chest, specifically a CT pulmonary angiogram (CTPA). The image displays the thoracic cage, lungs, mediastinum, and cardiac structures, notably demonstrating filling defects in the pulmonary arteries consistent with PE. The surrounding flowchart categorizes PE severity into three tiers: Low-Risk, Intermediate-Risk, and High-Risk. Low-Risk is defined by a Pulmonary Embolism Severity Index (PESI) score ≤ II and absence of Right Ventricular (RV) dilatation. High-Risk is identified by hemodynamic instability (circulatory shock, SBP < 90 mmHg, or cardiorespiratory arrest). The Intermediate-Risk section utilizes integrated biomarkers and scoring systems, including the BOVA score (focusing on SBP, troponin, RV dysfunction, and heart rate), the TELOS score (lactate, troponin, and RV dysfunction), and the SHIELD score (shock index, lactate, hypoxemia, and RV dysfunction). This educational visual illustrates how imaging findings of RV strain and clinical scores guide patient prognosis and triage.

This composite educational graphic illustrates the interpretability of a multimodal machine learning model for pulmonary embolism (PE) risk stratification. (a) A horizontal bar chart displays 'Feature Importances' from Electronic Health Record (EHR) data. Key clinical features contributing to model prediction include lactate levels, SaO2 at room air, PCO2, D-Dimer, calcium, patient age, and hypertension. (b) A 5x6 grid presents diagnostic interpretability through Grad-CAM visualizations. It pairs axial 2D slices from 3D Computed Tomography Pulmonary Angiograms (CTPA) with corresponding heatmaps. These heatmaps utilize a color gradient (red/yellow/green indicating high activation) to highlight anatomical regions prioritized by the model for classification. Predominant activation occurs within cardiothoracic structures, including the pulmonary trunk, right and left ventricles, aorta, inferior vena cava, and general heart area. This demonstrates the model's focus on physiologically relevant hemodynamic and vascular markers for determining PE severity and associated 30-day mortality risk.

This infographic presents misclassified examples from a multimodal classifier designed for Pulmonary Embolism (PE) risk stratification, comparing Electronic Health Record (EHR) features and 3D CT Pulmonary Angiography (CTPA) imaging. (a) Displays a false negative case featuring a bar chart of EHR feature importance (e.g., Pulmonary heart disease, Patients Complaint, CRP) and corresponding 3D Grad-CAM heatmaps overlaying the left ventricle and pulmonary artery. (b) Shows two false positive examples with their respective feature importance rankings (including Phosphorus, WBC, and Urea) and Grad-CAM visualizations. The heatmaps in these cases highlight the pulmonary trunk, aorta, right ventricle chamber, and areas indicating potential PE clots. This visual provides an educational comparison of how deep learning models utilize diverse data modalities to assess acute PE severity, illustrating anatomical regions of high diagnostic importance such as the heart chambers and major thoracic vessels. It serves as a study of model interpretability and the clinical relevance of multimodal diagnostic fusion.