Detection and Classification of Diabetic Macular Edema with a Desktop-Based Code-Free Machine Learning Tool

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic macular edema (DME) is a leading cause of vision loss in diabetic patients.
• Accurate classification of DME subtypes is crucial for effective treatment.
• Spectral-domain optical coherence tomography (SD-OCT) is a key imaging modality for DME diagnosis.

Purpose of the Study:

• To evaluate the effectiveness of the Lobe application, a code-free machine learning tool, for recognizing and classifying DME in SD-OCT images.
• To assess the performance of a machine learning model developed using Lobe for DME detection and subtyping.

Main Methods:

• Utilized a dataset of 695 SD-OCT images from patients with DME and 200 images from healthy controls.
• Employed the Lobe application with a pre-trained ResNet-50 V2 convolutional neural network for model development.
• Classified DME into diffuse retinal edema (DRE), cystoid macular edema (CME), and cystoid macular degeneration (CMD).

Main Results:

• The model achieved 99.28% sensitivity and 100% specificity for overall DME detection.
• Specificities for DRE, CME, and CMD were 98.57%, 99.29%, and 95.41%, respectively.
• Sensitivities for DRE, CME, and CMD were 87.80%, 96.43%, and 95.71%, respectively.

Conclusions:

• The Lobe application demonstrates high efficiency in recognizing and classifying DME from SD-OCT images.
• This code-free ML tool can be effectively utilized by ophthalmologists without requiring coding expertise.
• This study represents the first evaluation of Lobe with ophthalmological images, highlighting its potential in clinical practice.