A spatial correlation-guided deep fusion framework for multimodal lung cancer classification using CT imaging

Lung cancer is one of the main causes of death on the global level and thus needs precise and valid diagnostic methods. Traditional deep learning methods for lung cancer detection typically rely on single-modality inputs or naive fusion techniques, yet they cannot capture the intricate spatial correlations in medical data.

To address this drawback, this paper introduces a deep learning system based on spatial correlation for multimodal lung cancer classification. It proposes a mechanism called Spatial Correlation Mapping (SCM) to capture geometric and anatomical relationships among imaging data explicitly. This is combined with a multi-scale feature-extraction backbone and a correlation-guided fusion strategy to enable successful alignment and fusion of heterogeneous features without loss of spatial coherence. Extensive testing is conducted on benchmark lung cancer datasets to evaluate the proposed framework's efficiency.

The proposed model attains 98% accuracy and 100% recall on malignant tumors compared to baseline models, demonstrating the usefulness of the proposed concept of spatial correlation between features to enhance feature fusion and improve lung cancer diagnosis. The findings also reveal better precision, recall, and F1-score than a more traditional single-backbone and fusion-based approach.

Furthermore, the proposed model is computationally efficient enough to be competitive and applicable in real-world clinical settings. The results demonstrate the significance of spatial dependency modeling in enhancing multimodal analysis of medical images and offer a viable method for improving lung cancer diagnosis.