Multimodal Pulmonary Embolism Diagnosis Tool with Electrical Impedance Tomography and Electronic Health Records under...

Pulmonary embolism (PE) is a life-threatening condition frequently misdiagnosed or diagnosed with delays using current CT-based workflows, highlighting the need for accurate early detection tools. While recent deep learning-based multimodal frameworks for pulmonary disease diagnosis rely on large CT datasets, lung function assessments and electronic health records (EHRs), the high cost and limited accessibility of CT imaging pose challenges in resource-limited settings. Electrical impedance tomography (EIT), a non-invasive imaging modality without ionizing radiation, offers a competitive alternative for community-based screening of at-risk populations. This study proposes a deep learning multimodal framework that integrates lung EIT data with EHRs, grouped by availability, to quantify PE risk. Using a large CT-EHR dataset and simulated EIT data, we trained a multimodal deep learning model combining EIT and EHR inputs, which improved PE risk prediction compared to an EHR-only model. The EIT-based prediction achieved an 0.824 Area Under the Curve (AUC), with the multimodal framework reaching the AUC of 0.898. Furthermore, the imaging model outperformed the EHR model under conditions of scarce information, demonstrating its potential as a diagnostic tool in resource-constrained settings. To our knowledge, this study represents the first integration of EIT imaging with EHR datasets categorized by availability, highlighting the clinical potential of EIT across diverse data availability scenarios.Clinical Relevance-Our findings revealed the potential of EIT-based multimodal deep learning to achieve accurate and cost-effective PE risk prediction in clinical settings.