Deep learning and machine learning in CT-based COPD diagnosis: Systematic review and meta-analysis

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonology

Background:

• Chronic obstructive pulmonary disease (COPD) is a major global health challenge.
• Artificial intelligence (AI) shows promise in improving COPD diagnosis through CT imaging.
• Comprehensive evidence on the diagnostic performance of AI models for COPD is still needed.

Purpose of the Study:

• To quantitatively analyze the diagnostic performance of AI models in CT images for COPD.
• To provide evidence for the development of AI-driven COPD diagnostic tools.
• To compare the efficacy of different AI approaches in COPD diagnosis.

Main Methods:

• Systematic literature search of PubMed, Cochrane Library, Web of Science, and Embase up to September 1, 2024.
• Quality assessment of included studies using the QUADAS-2 tool.
• Meta-analysis of sensitivity, specificity, and area under the curve (AUC) using Stata18, RevMan 5.4, and Meta-Disc 1.4, including SROC curve plotting.

Main Results:

• Meta-analysis included 15 studies with 22,817 patients, evaluating deep learning (DL), machine learning (ML), and DL with multiple-instance learning (MIL) models.
• Pooled sensitivity was 86% (95% CI 78-91%), specificity 87% (95% CI 83-91%), and AUC 93% (95% CI 90-95%) for all AI models.
• No significant difference in diagnostic efficacy was found between DL and ML models; DL with MIL showed a trend towards improved performance, though not statistically significant.

Conclusions:

• Both DL and ML models demonstrate high accuracy for diagnosing COPD from CT images.
• There is no significant difference in diagnostic efficacy between DL and ML models.
• The multiple-instance learning (MIL) mechanism may enhance the performance of DL models in COPD diagnosis.