Lung Cancer Image Recognition Method Based on CMT Hybrid Model: Integrating Local Perception and Global Modeling

Lung cancer, a high-mortality malignancy, suffers from delayed primary diagnosis. To address limitations in traditional pathological diagnosisspecifically, insufficient local perceptual ability in medical image analysis and the high computational load of classical Transformerswe propose CMT, a hybrid CNN-Transformer model. CMT employs a convolutional encoder to extract multi-scale local features from input images. These features are transformed into global representations via Cross-Scale Feature Aggregation (CSFA) and processed by a Transformer decoder for final classification. The model is optimized using a weighted combination of cross-entropy and Dice loss functions, enhancing both accuracy and localization capability. Evaluated on the TCIA dataset, CMT achieved an accuracy of 92.8%, outperforming comparative methods.