Heart failure causes blood pressure to elevate, which in turn causes the fluid to leak from the blood vessels in the lungs into pleural space. The excess fluid in the lungs is called pulmonary edema. Knowing the severity of pulmonary edema is critical when making clinical decisions for heart failure patients, because it is the most direct symptom of heart failure and the treatment depends on the severity of the disease. Chest x-rays are commonly used to visualize the severity of pulmonary edema. Accurate grading of the pulmonary edema is challenging. Physicians rely on their subjective judgements on current patient status by observing the patients at the bedside when making treatment plans. Later, the radiology report captures the radiologist’s impression of the edema severity in a form of unstructured text, precluding quantitative analysis to improve prediction of the individual patient’s trajectory or to understand how a choice of a treatment strategy affects outcomes.
This project aims to develop machine learning algorithms to automatically quantify the severity of pulmonary edema from chest x-rays on a continuous scale. The resulting assessment can be used for visualization of patient recovery trajectories in prior episodes of heart failure to support physicians with a data-driven approach to treating patients. Furthermore, the resulting model will provide a quantitative phenotype to be used in large-scale population analysis of responses to different heart failure treatments, for improving the understanding of the disease progression and accelerating drug development.
Representative chest x-rays of pulmonary edema: