Research
1)  Review of applied methods for diabetic retinopathy detection, focusing on machine learning models such as convolutional neural networks (CNN) and artificial neural networks (ANN). and hybrid models. Each AI algorithm will then be classified by algorithm type (e.g. CNN, ANN, SVM), according to specific tasks in the process of detecting diabetic retinopathy. In particular, models implemented in the CNN direction will be analyzed further and classified depending on the number of important attributes of the corresponding CNN architectures in each model.
2)  Proposing a method for FOV segmentation based on the idea of ​​histogram and thresholding. A comparison is made with 11 segmentation methods on the four most commonly used retinal representations according to three metrics: Jaccard, Matthews correlation coefficient (MCC), and balanced accuracy. All options have been tested on 7,000 images from eight of the largest databases available. Test images were generated by about 100 different ocular fundus cameras. According to the analysis of the experimental results, the following conclusions were made: 1) segmentation of the FOV region should be performed on the red channel; 2) to reduce the uneven illumination effect and select the FOV area more accurately, the luminance must be converted into a logarithmic representation; 3) FOV region masking can be performed by calculating the global threshold (e.g., Otsu method); 4) calculating the global binary threshold by analyzing the peaks of the luminance histogram does not offer any advantage over the Otsu method, but is more sensitive to FOV scaling.
Publication
The results of the project has published on 01 paper in SCIE journal, 01 paper in Proceedings of RINTI 2022.
Education
The project supported 01 Master student under the research direction of the topic.

Review of applied methods for diabetic retinopathy detection, focusing on machine learning models such as convolutional neural networks (CNN) and artificial neural networks (ANN). and hybrid models.
Proposing a method for FOV segmentation based on the idea of ​​histogram and thresholding. A comparison is made with 11 segmentation methods on the four most commonly used retinal representations according to three metrics: Jaccard, Matthews correlation coefficient (MCC), and balanced accuracy. All options have been tested on 7,000 images from eight of the largest databases available.

- Publications:
01 paper in SCIE journal
[[1] V. V. Starovoitov, Nguyen Nhu Son*, Yu. I. Golub, M. M. Lukashevich, Nguyen Long Giang, Hoang Thi Minh Chau, Le Hoang Son, A Universal Field-of-View Mask Segmentation Method on Retinal Images from Fundus Cameras, IEEE Access, in press (2024).
01 papers in Proceeding of International Conderences
[2] M. Lukashevich, Nguyen Nhu Son*, Nguyen Long Giang, Le Hoang Son, Hoang Thi Minh Chau, V. Starovoitov, A review of deep learning models for detecting diabetic retinopathy from retinal images, RINTI-2022, Minsk, Belarus, 17/11/2022, pp. 157-163..

- Products: Summary report on the results of the project.

- Education: The project supported 01 Master student under the research direction of the topic.