Optimizing Pulmonary Embolism detection through diverse UNET architectural variations

Abstract

A Pulmonary Embolism (PE) occurs when a blood clot, usually originating from the deep veins in the legs (deep vein thrombosis), travels to the lungs and becomes lodged in a blood vessel, obstructing blood flow. This can lead to serious complications, including damage to lung tissue, decreased oxygen levels in the blood, and in severe cases, death. Given the potentially life-threatening nature of PE, accurate and timely diagnosis is crucial. Computed Tomography Pulmonary Angiography (CTPA) is a highly specialized X-ray technique commonly used by medical professionals to detect and precisely locate PE. CTPA involves injecting a contrast dye into the bloodstream and then taking detailed images of the pulmonary arteries using computed tomography (CT) scanning technology. These images allow physicians to visualize any blockages or abnormalities in the blood vessels of the lungs. In recent years, machine learning and deep learning techniques have gained traction in medical imaging analysis, offering the potential to automate and enhance diagnostic processes. Among these methodologies, U-shaped encoder-decoder architectures have shown promise for segmenting anatomical structures and abnormalities from medical images. This research aims to evaluate and compare the performance of several U-shaped networks, including UNET, UNET++, Residual UNET, ARUX, and Attention UNET, in accurately segmenting PE regions from CTPA images. By leveraging a publicly available PE challenge dataset, the study conducts comprehensive training, validation, and testing procedures, employing metrics such as the Dice Coefficient, Jaccard Similarity Index, and Sensitivity for meticulous evaluation. The findings of this study offer valuable insights into the efficacy and suitability of various deep learning architectures for PE segmentation. Ultimately, this research paves the way for enhanced diagnostic capabilities in clinical settings, potentially leading to improved patient outcomes and reduced mortality rates associated with this life-threatening condition.