Posters

Development of ophthalmological image database and intelligent computational methods for supporting medical diagnosis

Poster Details

First Author: C.Damtsi GREECE

Co Author(s):                        

Abstract Details

Purpose:

This study presents the creation of an ophthalmological image database from the fundus of the eye and the contribution of this database for the development of an artificial intelligence application that will recognize pathological cases in order to provide additional support for medical diagnosis. The whole process is fully in line with the requirements of the General Data Policy Regulations (GDPR).

Setting:

OPHTHALMICA, Institute of Ophthalmology & Microsurgery, Thessaloniki, Greece.

Methods:

1st phase: • 1682 fundus images were collected. • 718 normal \ 964 abnormal from 738 individuals. • Initial resolution 3680 x 3288, in jpeg format. • The images with pathological features were divided into 8 individual classes. • Each class includes sub-categories (Most commonly age-related macular degeneration and diabetic retinopathy). 2nd phase: • reduce of resolution and inserting images into the application • The network that was developed was LB-FCN, a fully convolutional neural network that separates the images into normal - abnormal with quite accurate results. •The database was edited in another network, MobileNet-v2 for comparative reasons

Results:

• For the LB-FCN CNN the AUC was 94.11% ±1.34% and the Accuracy was 95.03% ±1.41% • For the MobileNet-v2 CCN the AUC was 92.58% ± 1.57% and the accuracy was 93.44% ±1.79%

Conclusions:

Artificial intelligence and deep-learning algorithms based on convolutional neural networks (CNNs) are increasingly being applied to medical diagnosis, by giving new opportunities for health promotion. However, the final diagnosis must be fulfilled by a qualified physician, considering other relevant factors.

Financial Disclosure:

None