Lung Nodule Diagnosis Using New Astrocyte-Illumined Neural Networks with Transfer Learning

Lung cancer has a high mortality rate, and early diagnosis can significantly improve patient survival probabilities. While neural network-based methods have shown promise in diagnosing lung nodules at earlier stages, their effectiveness is limited by the small scale of available medical datasets, leading to high training costs and relatively poor performance. To address these challenges, we employ transfer learning to build a fine-tuned AstroNet to diagnose lung nodules. The proposed method is an astrocyte-illumined new artificial neural network, which has been trained on a large-scale dataset of normal images. We apply a partial fine-tuning strategy to enhance the learning efficiency of AstroNet. Experimental results demonstrate that fine-tuned AstroNet outperforms traditional classification networks in the lung nodule classification task. Visualization analysis further confirms that fine-tuned AstroNet captures the unique features of lung nodules more comprehensively. The low time cost of the proposed method also has a good prospect in the clinical application.