Performance of Deep-Learning Solutions on Lung Nodule Malignancy Classification: A Systematic Review

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Computational Pathology

Background:

• Computer technology aids lung nodule diagnosis.
• Deep learning surpasses traditional machine learning in image processing for nodule diagnosis by eliminating manual feature extraction.
• Deep learning methods offer improved accuracy in lung nodule diagnosis.

Purpose of the Study:

• To investigate the efficacy of deep learning approaches in classifying lung nodule malignancy.
• To systematically review and evaluate the performance of deep learning models for lung nodule malignancy classification.

Main Methods:

• Systematic literature search of PubMed and ISI Web of Science databases.
• Selection of studies employing deep learning for lung nodule malignancy classification or prediction.
• Data extraction and analysis using SAS version 9.4 and Microsoft Excel 2010.

Main Results:

• Sixteen studies were included in the analysis.
• Deep learning models, including Convolutional Neural Networks (CNNs), Autoencoders (AEs), and Deep Belief Networks (DBNs), were utilized.
• Deep learning models demonstrated high performance with Area Under the Curve (AUC) typically exceeding 90%.

Conclusions:

• Deep learning shows distinct advantages in pulmonary nodule diagnosis and forecasting.
• This analysis highlights recent advancements in deep learning for lung nodule detection.
• Further research is warranted to address the drawbacks associated with deep learning models in this field.