RESOURCE-LIMITED SETTINGS LOW-COST AI-ASSISTED CANCER SCREENING AND ICU TRIAGE MODELS FOR RESOURCE-CONSTRAINED HEALTHCARE SYSTEMS

Abstract

The lack of cancer specialists, diagnostic infrastructure, access to imaging, lab capacity, and critical care beds present significant challenges for resource constrained healthcare systems to provide timely cancer screening and effective intensive care unit triage. In these contexts, AI could provide scalable low-cost assistance for early risk identification, prioritization and clinical decision making. This study seeks to propose a low cost AI based framework for cancer screening and triage in the Intensive Care Unit built using common clinical, demographic, symptom-based, laboratory and basic imaging features. The framework was developed to be efficient and feasible to implement in primary care clinics, district hospitals and low-resource oncology and emergency care settings. Several machine learning models were tested for the prediction of cancer screening priority, admission to the intensive care unit, ventilator use, mortality risk, and referral urgency.Multiple machine learning models were tested, such as Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, XGBoost, as well as lightweight neural network classifiers. The accuracy, sensitivity, specificity, precision, F1-score, calibration performance and area under the receiver operating characteristic curve were used to evaluate the model performance. Proposed low-cost AI assisted model had excellent prediction capability with good interpretability and operability. The following factors were identified as important: age, symptom duration, smoking history, weight loss, oxygen saturation, respiratory rate, blood pressure instability, hemoglobin level, inflammatory markers, basic imaging abnormality, comorbidity burden and performance status. The results indicate that low-cost AI systems could be used to aid a earlier cancer diagnosis and to streamline and enhance the triage process in ICUs where health resources are scarce. This can help frontline health-care providers prioritize high-risk patients, shorten the diagnostic delay, aid in referrals, and maximize limited critical care resources.