Parameterized Gompertz-Guided Morphological AutoEncoder for Predicting Pulmonary Nodule Growth.

Area of Science:

• Medical imaging analysis
• Computational pathology
• Artificial intelligence in oncology

Background:

• Pulmonary nodule growth rate is a critical indicator for cancerous diagnosis.
• Monitoring dynamic nodule progression is essential for effective pulmonary nodule management.
• Lack of temporal datasets hinders research in nodule growth prediction.

Purpose of the Study:

• To develop a model for predicting pulmonary nodule growth for improved cancer diagnosis.
• To create a parameterized Gompertz-guided morphological autoencoder (GM-AE) for predicting future nodule appearances.
• To quantitatively predict pulmonary nodule growth rates using computed tomography (CT) scans.

Main Methods:

• Organized and published the temporal dataset NLSTt with consecutive CT scans.
• Developed a visual learner for qualitative prediction of nodule growth.
• Proposed a parameterized Gompertz-guided morphological autoencoder (GM-AE) for quantitative growth prediction.
• Utilized the Gompertz model to predict future nodule mass and volume, guiding shape and texture generation.
• Implemented a dual-branch autoencoder for shape-aware and textural-aware representation learning.

Main Results:

• The GM-AE model demonstrated superior performance compared to existing methods on the NLSTt dataset.
• Experiment results confirmed the descriptive power of the learnable Gompertz function in capturing inter-subject variability of nodule growth rates.
• The GM-AE model showed generalizability and practicality when evaluated on an in-house dataset.

Conclusions:

• The proposed GM-AE model effectively predicts future pulmonary nodule morphology and growth rates.
• The Gompertz function parameterization provides valuable insights into individual nodule growth patterns.
• The publicly available NLSTt dataset and GM-AE code will facilitate further research in pulmonary nodule analysis.