ADGAN: Attribute-Driven Generative Adversarial Network for Synthesis and Multiclass Classification of Pulmonary...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer is a leading cause of cancer deaths globally.
• Early detection of pulmonary nodules via computed tomography (CT) scans significantly improves survival rates.
• Accurate classification of pulmonary nodules is crucial for effective treatment planning.

Purpose of the Study:

• To propose an attribute-driven Generative Adversarial Network (ADGAN) for the synthesis and multiclass classification of pulmonary nodules.
• To enhance the generation mechanism using a self-attention U-Net (SaUN) architecture.
• To improve the accuracy of pulmonary nodule detection and classification in CT scans.

Main Methods:

• Developed an ADGAN model incorporating a self-attention U-Net (SaUN) generator with self-attention attribute (SaAM) and spatial (SaSM) modules.
• Utilized reconstruction loss and attention localization loss (AL) to generate attention maps for nodule regions.
• Implemented an adversarial loss with KL divergence regularization to prevent image resemblance and a discriminator for multiclass classification.

Main Results:

• The proposed ADGAN model demonstrated promising classification accuracy on the LIDC-IDRI and LUNGX datasets.
• The self-attention mechanisms in the SaUN architecture improved nodule synthesis and localization.
• Achieved competitive performance compared to existing state-of-the-art methods in pulmonary nodule classification.

Conclusions:

• The ADGAN approach offers an effective method for synthesizing and classifying pulmonary nodules.
• This technique has the potential to aid in the early diagnosis and treatment planning of lung cancer.
• The proposed model shows significant promise for advancing computer-aided diagnosis in medical imaging.