Deep learning in pulmonary nodule detection and segmentation: a systematic review.

Area of Science:

• Artificial intelligence in medical imaging
• Radiology and medical diagnostics
• Computational pathology

Background:

• Accurate lung nodule detection and segmentation on CT scans are crucial for early lung cancer diagnosis and treatment.
• Existing literature has methodological gaps and biases in deep learning approaches for pulmonary nodule analysis.

Approach:

• A systematic review of studies published between 2019-2023 was conducted using major scientific databases.
• Quality assessment of diagnostic accuracy and AI in medical imaging checklists were employed to evaluate bias.
• Model performance, data sources, and task focus were extracted from nine included studies.

Key Points:

• Deep learning techniques, primarily Convolutional Neural Networks, are increasingly used for lung nodule detection and segmentation.
• Public datasets like LIDC-IDRI and LUNA16 are predominantly utilized.
• Performance metrics such as sensitivity and Dice coefficient are commonly reported.

Conclusions:

• Deep learning holds significant potential for autonomous lung nodule detection and segmentation.
• Standardized data processing, code/data sharing, and external validation are essential for enhancing clinical utility.
• Addressing methodological shortcomings and ensuring transparency are critical for future research and clinical application.