Intelligent Diagnostic Model for Pulmonary Disease based on Multi-modal Data Integration Strategy of Chest CT images

Due to the insidious onset and biological heterogeneity of non-small cell lung cancer (NSCLC), early detection is necessary but challenging. Accurate identification of tumor stage and histological sub-types in NSCLC is a critical clinical challenge. This study proposed an intelligent diagnostic method for NSCLC tumor stages classification and tumor subtypes identification based on multi-modal data integration strategy of chest CT images. We analyzed the chest CT images of a publicly available set of 422 patients from the MAASTRO NSCLC collection (Lung1). Compared with other 11 different models, the proposed model for tumor stages classification has the best performance, achieving an accuracy of 89%, a precision of 86%, a recall of 88%, and an F1-score of 85%. The proposed model for tumor sub-types identification achieved an accuracy of 84%, a precision of 79%, a recall of 84%, and an F1-score of 79%. This method can help clinicians in identifying high-risk individuals at an early stage and facilitating timely treatment, thereby supporting effective triage of patients upon admission.