Automatic Pulmonary Nodule Detection Applying Deep Learning or Machine Learning Algorithms to the LIDC-IDRI Database:...

Area of Science:

• Radiology
• Medical Imaging
• Computer Science

Background:

• Lung nodule detection in thoracic CT scans is crucial for early diagnosis.
• Machine learning (ML) and deep learning (DL) offer advanced tools for image analysis.
• The Lung Image Database Consortium Image Collection (LIDC-IDRI) is a key resource for developing and validating these algorithms.

Purpose of the Study:

• To systematically review literature on ML algorithms applied to the LIDC-IDRI database for lung nodule detection.
• To provide an overview of algorithm architectures and performance in optimizing lung nodule detection.
• To assess the accuracy, sensitivity, and specificity of ML and DL in analyzing thoracic CT scans.

Main Methods:

• Systematic literature review adhering to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.
• Inclusion of original research articles focusing on ML algorithms applied to the LIDC-IDRI database.
• Categorization of included articles based on algorithm architecture (feature-based vs. deep learning).

Main Results:

• A total of 41 articles were included after an initial search of 1972 publications.
• Feature-based ML algorithms predominantly achieved accuracy greater than 90%.
• Deep learning (DL) algorithms demonstrated accuracy ranging from 82.2% to 97.6%.

Conclusions:

• ML and DL algorithms show high accuracy, sensitivity, and specificity for lung nodule detection in CT scans.
• Feature-based ML algorithms generally outperform DL methods in terms of accuracy.
• There is a lack of consensus regarding standardized methods for evaluating the efficiency of ML algorithms in this domain.