Hybrid AI System Improves Early Lung Cancer Detection on CT

Early lung cancer often presents as tiny nodules on computed tomography (CT) that are difficult to distinguish from normal tissue, leading to delayed diagnoses. Missed or late findings limit timely intervention and worsen survival. Clinicians also face heavy imaging workloads that increase the risk of oversight. To help address this challenge, researchers at Kaunas University of Technology (Kaunas, Lithuania) have developed an artificial intelligence system designed to support earlier lung cancer detection directly from CT scans.

The innovation is a hybrid deep learning model that integrates a convolutional neural network (CNN) and a transformer to analyze CT images. One component is optimized for local textures and minute parenchymal findings. The other captures global anatomical context across the full lung field. By learning both views in a single pass, the system mirrors how radiologists alternate between zoomed-in and whole-chest assessments without requiring manual switching.

Image: The hybrid deep learning model that integrates a convolutional neural network and a transformer to analyze CT images (photo courtesy of Shutterstock)

The model was trained on CT scans from healthy individuals and patients with lung cancer, learning patterns that differentiate normal, benign, and malignant cases. This dual-view strategy addresses a common limitation of prior systems that emphasized either fine detail or broader structure but not both simultaneously. In practice, the approach enables concurrent detection of subtle lesions while maintaining awareness of their relationship to surrounding anatomy.

In evaluations reported by the research team, the system achieved accuracy exceeding 96% and outperformed existing approaches, with stable performance across tests. It is designed to reduce missed cancers and lower false alarms that can prompt unnecessary procedures. The findings were published in Scientific Reports on March 17, 2026. Next steps include clinical validation in hospital settings, testing across diverse scanners and imaging protocols, and integration into clinical workflows. The same framework may extend to brain tumors, breast cancer, and eye diseases.

"One part of the model focuses on small details, such as tiny spots or textures in the lungs, while another looks at the overall image and understands the bigger context. You can think of it as having a magnifying glass and a full view of the scan at the same time," said Inzamam Mashood Nasir, researcher at Kaunas University of Technology and one of the system's developers.

Related Links
Kaunas University of Technology