Multi-task learning for predicting pulmonary nodule growth and follow-up volume.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in radiology
• Deep learning for medical diagnosis

Background:

• Pulmonary nodules require accurate growth prediction for effective clinical surveillance.
• Existing methods may lack precision in modeling complex nodule evolution.
• Deep learning offers potential for enhanced predictive capabilities in nodule management.

Purpose of the Study:

• To develop an end-to-end deep learning framework for predicting pulmonary nodule growth.
• To jointly model nodule segmentation and visual follow-up image synthesis using multi-task learning.
• To improve predictive accuracy and clinical applicability by decoupling nodule growth into deformation and texture evolution.

Main Methods:

• Introduced MT-NoGNet, a dual-task network for simultaneous deformation-texture modeling of pulmonary nodules.
• Employed a shared encoder with two decoders: a spatial transformer for volume change estimation and a texture predictor.
• Utilized a cross-task attention mechanism to ensure consistency between morphological expansion and internal density evolution.

Main Results:

• Evaluated on longitudinal CT scans from 246 patients.
• Achieved a predicted peak signal-to-noise ratio (PSNR) of 44.30.
• Obtained a structural similarity index (SSIM) of 0.7776 and a Dice similarity coefficient (DSC) of 0.7823.

Conclusions:

• Multi-task learning of deformation-texture features significantly enhances pulmonary nodule growth prediction accuracy.
• The model provides radiologists with interpretable visualizations of progression patterns.
• Demonstrated substantial potential for optimizing clinical surveillance protocols for pulmonary nodules.