Related Experiment Video
Updated: May 8, 2025
07:53
Author Spotlight: Advancing 3D Modeling for Enhanced Diagnosis and Treatment of Pulmonary Nodules in Early-Stage Lung Cancer
Published on: October 13, 2023
1.3K
Fusion-driven semi-supervised learning-based lung nodules classification with dual-discriminator and dual-generator
Ahmed Saihood1, Wijdan Rashid Abdulhussien1, Laith Alzubaid2,3,4
1College of Computer Science and Mathematics, University of Thi-Qar, Thi Qar, Iraq.
BMC Medical Informatics and Decision Making
|December 24, 2024
Summary
A novel dual-generator, dual-discriminator generative adversarial network (DDDG-GAN) improves semi-supervised lung nodule classification by preventing mode collapse and enhancing generalizability. This approach shows superior performance on diverse datasets, aiding in accurate lung cancer diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Accurate lung nodule detection and classification are vital for lung cancer diagnosis and treatment.
- Existing models struggle with mode collapse and poor generalizability, limiting their effectiveness.
- Semi-supervised learning offers a promising avenue for improving lung nodule classification.
Purpose of the Study:
- To propose a novel generative adversarial network (GAN) architecture, DDDG-GAN, for semi-supervised lung nodule classification.
- To address mode collapse and enhance model generalizability in lung nodule classification tasks.
- To evaluate the performance of DDDG-GAN on benchmark datasets and compare it with existing methods.
Main Methods:
- Developed a DDDG-GAN model featuring dual generators (benign/malignant) and dual discriminators.
- Employed feature fusion techniques to improve discriminatory power between nodule classes.
- Evaluated the model in two scenarios: within-dataset (LIDC-IDRI) and cross-dataset (LIDC-IDRI to LUNA16/LUNGx).
Main Results:
- DDDG-GAN achieved high performance on LIDC-IDRI (e.g., 92.56% accuracy).
- Demonstrated robust cross-dataset performance on unseen LUNA16 (72.6% accuracy) and LUNGx (71.23% accuracy) datasets.
- Outperformed state-of-the-art semi-supervised learning approaches in both evaluation scenarios.
Conclusions:
- The DDDG-GAN effectively mitigates mode collapse and enhances generalizability in lung nodule classification.
- The model shows significant potential for improving diagnostic accuracy in clinical settings.
- DDDG-GAN offers a robust solution for semi-supervised learning in medical image analysis.
