TWO-STAGE PREDICTION FOR PULMONARY NODULE GROWTH BASED ON GROWTH KINETICS MODEL

Pulmonary nodules, as common lung lesions, have a potential risk of malignant transformation, which may turn into lung cancer. In medical practice, early diagnosis and timely intervention are considered as key strategies for lung cancer treatment, which are of great significance for improving the effectiveness of treatment. With the continuous advancement of medical technology, computer-aided diagnosis technology has demonstrated its indispensable role in the early screening and treatment of lung cancer. However, the current studies focus on onetime imaging data, which neglect the pulmonary nodule growth over time. In this paper, we propose a two-stage pulmonary nodule growth prediction network based on a growth kinetics model. The network consists of a mass and volume prediction sub-network and a shape and texture generation sub-network. The mass and volume prediction sub-network comprises a vision learner with ViT as the backbone network, which receives processed image data as input and obtains factor values for predicting nodule growth rates through a parametrized Gompertz function. The shape and texture generation sub-network utilizes the obtained growth rates of volume and mass to generate high-quality and realistic visual appearances of pulmonary nodules. The experimental results show that the Dice Similarity Coefficient (DSC) and Positive Predictive Value (PPV) are significantly improved by 2.35% and 3.11%, respectively, compared to the baseline model.