A Deep Learning Model for Screening Multiple Abnormal Findings in Ophthalmic Ultrasonography (With Video)

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Ophthalmic ultrasound is crucial for diagnosing posterior segment diseases.
• Current diagnostic methods can be time-consuming and require expert interpretation.
• Need for efficient and accurate screening tools for conditions like retinal detachment (RD), vitreous detachment (VD), and vitreous hemorrhage (VH).

Purpose of the Study:

• To develop a deep learning system for real-time screening of RD, VD, and VH in ophthalmic ultrasound.
• To assess the accuracy and efficiency of the AI system in identifying multiple abnormal findings.

Main Methods:

• A deep convolutional neural network was employed.
• Trained on a large dataset of 3580 images for classification and 941 for segmentation.
• Validated on external datasets and tested in real-time video analysis.

Main Results:

• The model achieved high accuracy in recognizing normal and abnormal findings (0.90-0.94).
• Demonstrated superior or comparable performance to experts in classifying RD, VD, and VH (0.73 accuracy).
• Significantly improved trainee recognition accuracy when used as an assistive tool (0.84 to 0.94).

Conclusions:

• The deep learning model serves as an effective screening tool for RD, VD, and VH.
• The system shows potential as a valuable aid in ophthalmology training.
• AI integration can enhance the accuracy and efficiency of ophthalmic ultrasound interpretation.