AI-Assisted Early Screening of Lung Cancer: Evaluation of Deep Learning Models based on CT Images

Aiming at the problems of low efficiency and insufficient accuracy of manual reading in early CT image screening of lung cancer, this paper proposes an original multi-scale attention fusion convolutional neural network (MSA-FCNN). The model extracts multi-level features based on the improved residual network, obtains information of different scales through the multi-scale feature generation module, and combines the spatial and channel attention mechanisms to highlight the key lesion features, and finally achieves benign/malignant binary classification. The experiment was conducted on the LIDC-IDRI dataset and a private dataset. The results showed that the accuracy of MSA-FCNN on the LIDC-IDRI dataset was 96.8%, the sensitivity was 95.2%, the specificity was 97.5%, and the AUC was 0.983; the accuracy was 95.3%, the sensitivity was 93.8%, the specificity was 96.7%, and the AUC was 0.976 on the private dataset, which were significantly better than mainstream algorithms such as ResNet-50. This model provides an efficient and reliable auxiliary tool for early screening of lung cancer.