Classification of Stages 1,2,3 and Preplus, Plus disease of ROP using MultiCNN_LSTM classifier.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Retinopathy of prematurity (ROP) is a leading cause of blindness in premature infants.
• Accurate staging and identification of Plus disease are critical for timely ROP treatment.
• Current diagnostic methods rely on manual retinal image examination, which can be subjective and time-consuming.

Purpose of the Study:

• To develop and evaluate an explainable automated system for ROP screening.
• To improve the accuracy of classifying ROP stages (1-3) and detecting Pre-plus/Plus disease.
• To leverage deep learning models for enhanced ROP diagnosis from retinal fundus images.

Main Methods:

• Utilized multiple Convolutional Neural Networks (CNNs) for feature extraction from retinal images.
• Employed Long Short-Term Memory (LSTM) networks for image classification.
• Developed and used the Cropped STAGE and HVDROPDB-PLUS datasets, including RetCam and Neo images.

Main Results:

• The proposed MultiCNN-LSTM networks demonstrated superior performance compared to individual CNNs and CNN-LSTM models.
• Achieved higher accuracy and F1 scores in classifying ROP stages and identifying Plus disease.
• The system provides explainable classifications for ROP screening.

Conclusions:

• The developed explainable AI system shows significant potential for accurate and efficient ROP screening.
• This automated approach can aid ophthalmologists in early detection and treatment of ROP.
• Further validation on larger datasets is warranted to confirm clinical utility.