Semi-Supervised Lung Cancer Detection Algorithm Based on the Teacher-Student Framework

Lung cancer is one of the leading causes of cancer deaths worldwide. Due to the scarcity of labeled lung tumor data and the high cost of annotation, this paper proposes a semi-supervised lung cancer detection algorithm based on the teacher-student framework. Firstly, the teacher-student framework is introduced as the foundation of the semi-supervised detection, enabling improved detection performance with limited labeled lung cancer data through the collaboration of teacher and student models. Secondly, an MCIB module is proposed, which integrates multi-scale and SE channel attention to enhance the extraction of multi-scale information. Finally, a GLA module is introduced, which combines global and local attention to improve the model's contextual understanding and reduce noise interference. The proposed algorithm is evaluated on the Lung-PET-CT-Dx dataset, achieving a Precision of 87.6% and a Recall of 84.1% when using 10% labeled and 90% unlabeled data, surpassing the Efficient Teacher algorithm by 2.5% and 4.7%, respectively. Experimental results demonstrate that the proposed algorithm achieves high detection accuracy even with limited labeled data.