Identifying Medical Diagnoses and Treatable Diseases by Image-Based Deep Learning

Area of Science:

• Artificial Intelligence in Medicine
• Medical Imaging Analysis
• Ophthalmology
• Radiology

Background:

• Clinical decision support algorithms in medical imaging often struggle with reliability and interpretability.
• Accurate and timely diagnosis of blinding retinal diseases and pediatric pneumonia is crucial for effective treatment and improved patient outcomes.

Purpose of the Study:

• To develop and validate a deep-learning framework for diagnosing common treatable blinding retinal diseases and pediatric pneumonia.
• To enhance diagnostic transparency and interpretability in AI-driven medical imaging analysis.
• To demonstrate the general applicability and efficiency of the AI system for expedited diagnosis and referral.

Main Methods:

• Utilized a deep-learning framework with transfer learning for efficient neural network training on optical coherence tomography (OCT) images.
• Applied the framework to classify age-related macular degeneration and diabetic macular edema, comparing performance to human experts.
• Validated the AI system's generalizability by applying it to chest X-ray images for pediatric pneumonia diagnosis.
• Incorporated methods to highlight regions recognized by the neural network for interpretable diagnosis.

Main Results:

• Achieved diagnostic performance comparable to human experts in classifying age-related macular degeneration and diabetic macular edema from OCT images.
• Demonstrated the AI system's ability to accurately diagnose pediatric pneumonia using chest X-ray images.
• Provided a more transparent and interpretable diagnostic process by visualizing the AI's decision-making regions.

Conclusions:

• The developed deep-learning framework offers a reliable and interpretable AI solution for medical imaging diagnosis.
• This AI tool has the potential to expedite diagnosis and referral for treatable conditions like retinal diseases and pneumonia.
• Earlier diagnosis and treatment facilitated by this AI system can lead to improved clinical outcomes.