Lung cancer, particularly non-small cell lung cancer (NSCLC), remains the most common and aggressive type, with limited survival outcomes. While CT imaging is the standard for staging and monitoring, current manual and subjective assessments are time-consuming and inconsistent.
Our study introduces LLSB-CFPR, a novel deep learning pipeline designed to automatically segment multiple lung cancer lesions across diverse, real-world CT datasets. Unlike conventional methods that only focus on primary tumors, this approach captures all detectable lesions, enabling a comprehensive view of disease burden that is crucial for prognosis, treatment planning, and monitoring.
By addressing real-world imaging variability and extending beyond single-lesion segmentation, this work represents a significant step toward reliable, automated tools that can support clinicians in routine lung cancer care.
Xavier Rafael-Palou; Ana Jimenez-Pastor; Luis Martí-Bonmatí; Carlos F. Muñoz-Nuñez; Mario Laudazi; Ángel Alberich-Bayarri