Human-centered CT-based CADx Tools for Traumatic Torso Hemorrhage

点击翻译按钮获取中文摘要

Trauma is responsible for 180,000 deaths annually in the United States and accounts for 59% of deaths in the population younger than 45 years. 86% of preventable deaths are related to sequalae of massive torso hemorrhage. Rapid precision diagnostic tools are needed to triage patients for early activation of massive transfusion protocols and urgent surgical or angiographic hemostatic intervention to circumvent the vicious cycle of acidosis, coagulopathy, hypothermia, and death resulting from exsanguination. Whole-body CT angiography (WBCTA) is the workhorse screening and surgical planning modality for torso hemorrhage. Lethal but preventable hemorrhage typically arises from pelvic fractures and solid organ lacerations, manifesting on WBCTA as foci of contrast extravasation and pooled cavitary hemorrhage (e.g., pelvic hematoma, hemothorax, or hemoperitoneum). Rapid assessment of WBCTA can result in earlier intervention, with associated survival benefit, but reader fatigue, study volume, reading room distractions, and injuries involving multiple body regions remain sources of diagnostic error and interpretation delays. Assessment of organ injury severity, pelvic fracture severity, and overall hemorrhage burden remains reader dependent and subjective. Clinical tools including the Shock Index have many confounders that impede forecasting of actionable hemorrhage- related outcomes. Automated WBCTA computer aided diagnosis (CADx) tools that detect bleeding pelvic fractures and organ lacerations, classify severity grade, and deliver precise voxelwise volumetric measurements of multicavitary hemorrhage burden will greatly accelerate and standardize image analysis, reduce turnaround time for reporting of critical results, improve the accuracy and objectivity of clinical decision making, and ultimately reduce time to life-saving hemorrhage control interventions. To capitalize on the benefits of automated point-of-care CT-based CADx tools in the fast-paced, and safety critical trauma care setting, such tools must be rapid, accurate, generalizable, and elicit a high level of end-user trust. To minimize bias, ensure clinical utility, and maximize robustness for turn-key deployment in future multicenter clinical trials, the tools must scale to large diverse populations, and achieve human factors engineering goals established through expert target user input. Our team will bring to bear combined technical and clinical expertise in trauma radiology, medical image processing, and human-centered software design to create an orchestrated suite of rapid, accurate, clinically relevant, and user-centered CADx tools for torso hemorrhage. In Aim 1, we will curate and annotate a uniquely large dataset of consecutively selected admission trauma WBCTAs. In Aim 2, we will use this big data approach and human-centered design principles to develop a suite of interactive high-trust CADx tools. In Aim 3, we will assess generalizability with a large out-of-sample dataset and assess user acceptance with simulated deployment. The work will result in rapid, robust, and human-centered CADx tools for detection, precision diagnostics, and personalized decision support for hemorrhage-control interventions.