Automated malarial retinopathy detection using transfer learning and multi-camera retinal images

Area of Science:

• Ophthalmology
• Infectious Diseases
• Artificial Intelligence in Medicine

Background:

• Cerebral malaria (CM) is a severe, fatal condition primarily affecting children under five in Sub-Saharan Africa and Asia.
• Malarial retinopathy (MR), characterized by specific retinal lesions, is a key indicator for CM detection.
• Over-diagnosis of CM due to overlapping symptoms with other conditions leads to delayed or incorrect treatment, increasing mortality and neurological disability.

Purpose of the Study:

• To develop a low-cost, high-specificity diagnostic technique for cerebral malaria (CM).
• To leverage transfer learning (TL) for accurate detection of malarial retinopathy (MR) from retinal images.

Main Methods:

• Utilized transfer learning (TL) models pre-trained on large datasets to select high-quality retinal images.
• Employed a secondary TL model to analyze selected retinal images for the presence of malarial retinopathy (MR).
• Tested the diagnostic approach using low-cost retinal cameras.

Main Results:

• The developed transfer learning (TL) method demonstrated high specificity in detecting cerebral malaria (CM) through malarial retinopathy (MR).
• Achieved an overall specificity of 96% for CM detection.
• The approach is suitable for implementation with affordable retinal imaging equipment.

Conclusions:

• Transfer learning (TL) offers a promising, cost-effective solution for accurate cerebral malaria (CM) diagnosis.
• This method can significantly improve diagnostic accuracy, reducing misdiagnosis and improving patient outcomes.
• The high specificity of this technique supports its potential for widespread use in endemic regions.