A machine learning approach for improved resource allocation in COVID-19 ICUs using HRCT scans and clinical data

The COVID-19 pandemic continues to pose a significant threat to global health and economies. While initial surges may have subsided, vigilance remains crucial. Improvised methods for managing COVID-19 intensive care units (ICUs) highlight the need for improved management strategies. Machine learning techniques offer promise in this fight by enabling better medical resource allocation, faster triage of potentially critically ill patients, and efficient treatment delivery. Overwhelmed ICU capacity and suboptimal mechanical ventilation configuration are key challenges in such situations. This study proposes a novel approach for automated diagnosis of lung damage severity and prognosis of COVID-19 patients using high-resolution computed tomography (HRCT) scans. We combined HRCT analysis with clinical laboratory data to develop a simple and rapid lung damage severity score. This score was then utilized to predict ICU admission and mortality risk with promising accuracy. The U-Net neural network was employed to identify lung regions damaged by COVID-19 from axial HRCT images. These results were subsequently fed into separate machine learning models (support vector machines and Naive Bayes) for patient outcome prediction. Image processing techniques were also implemented to analyze axial HRCT scans. These combined results were used to train an artificial neural network to diagnose treatment stages and predict final patient outcomes. We further addressed the crucial yet under-investigated question of optimal timing for ventilator configuration adjustments. The proposed solution achieved comparable or superior performance compared to existing methods. This study demonstrates the potential of Machine Learning techniques for improved COVID-19 patient management. The proposed method offers a robust and automated approach for lung damage severity assessment and patient outcome prediction, potentially aiding in better resource allocation and patient care decisions.