Diabetic Retinopathy Lesion Segmentation Method Based on Multi-Scale Attention and Lesion Perception

Algorithms, Journal Year: 2024, Volume and Issue: 17(4), P. 164 - 164

Published: April 19, 2024

The early diagnosis of diabetic retinopathy (DR) can effectively prevent irreversible vision loss and assist ophthalmologists in providing timely accurate treatment plans. However, the existing methods based on deep learning have a weak perception ability different scale information retinal fundus images, segmentation capability subtle lesions is also insufficient. This paper aims to address these issues proposes MLNet for DR lesion segmentation, which mainly consists Multi-Scale Attention Block (MSAB) Lesion Perception (LPB). MSAB designed capture multi-scale features while LPB perceives depth. In addition, novel function with tailored weight reduce influence imbalanced datasets algorithm. performance comparison between other state-of-the-art carried out DDR dataset DIARETDB1 dataset, achieves best results 51.81% mAUPR, 49.85% mDice, 37.19% mIoU 67.16% mAUPR 61.82% mDice dataset. generalization experiment IDRiD 59.54% among methods. show that has outstanding ability.

Language: Английский

---

A Miniaturized Large-Field Fundus Optical System Based on Aspheric Imaging and Non-Coaxial Illumination

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(14), P. 6216 - 6216

Published: July 17, 2024

Many diseases produce pathological changes in the fundus; analyzing retinopathy of fundus could help diagnose time. A camera is a medical imaging device that specializes taking images to hypertension, coronary heart disease, diabetes, and others. The optical system core part it. Nevertheless, conventional large not suitable for mobile examination follow-up use. So, it has been widely used institutions. In this paper, miniaturized based on aspheric technology non-coaxial illumination proposed. length only 34.6 mm, field view 50°, MTF curve greater than 0.2 at 100 lp/mm, which can resolve structure 5 um. adopts annular array avoid occlusion system. Our study effectively tackles pressing predicament miniaturization. This innovative paradigm harbors potential revolutionize image data acquisition, propelling diagnosis forward efficiently catering crucial applications, improving versatility examination, providing technical support intelligent

Language: Английский

---

Segmentation of Drusen in Retinal Images for the Detection of Diabetic Retinopathy

International Journal For Multidisciplinary Research, Journal Year: 2024, Volume and Issue: 6(2)

Published: March 9, 2024

Diabetic retinopathy (DR) is a most complicated eye disease that affects people with diabetes for an extended period. If necessary treatment not given at the right time, DR leads to severe loss of vision without prior symptoms. Therefore, patients are recommended undergo continuous screening early detection DR. In this paper, we proposed automated process detect lesion called Drusen in retinal images. related The presence does indicate Still, it used identify severity level patient and avoid misdetection rate other bright lesions (both exudates cotton wool spots). This method based on Background image approach inverse segmentation area affected by Inverse segment healthy areas regular texture rather than varying unhealthy areas. segmented compared original segmenting Drusen. involved 40 images from STARE database, producing better results accuracy processing time.

Language: Английский

---

Predicting Diabetic Retinopathy Severity with Deep Learning: A Survey of Fundus Image Analysis Technique

Published: April 12, 2024

Diabetic retinopathy (DR) is a significant complication of diabetes mellitus, impacting vision due to retinal abnormalities. Early detection and precise severity assessment are crucial for effective management. Leveraging deep learning techniques image preprocessing methods, this paper proposes comprehensive approach DR classification. Utilizing publicly available datasets like EyePACS, Messidor-2, APTOS, DDR, steps including Gaussian blurring data augmentation employed enhance quality address class imbalance. Wavelet decomposition used feature extraction capture multi-resolution information from fundus images. Transfer with ResNet variants, coupled regularization techniques, aids in model generalization. A modified ResNet50 architecture introduced, featuring custom fully connected layers additional convolutional improved extraction. The aims classify diseases into four levels: normal, mild, moderate, severe proliferative. survey aspect delves methods' effectiveness improving CNN performance medical analysis, specifically detection. applicability transfer imaging tasks, particularly DR, also explored. This study contributes advancing analysis diagnosis classification, addressing the critical need efficient management debilitating condition.

Language: Английский

---

Automatic Detection of Optic Disc and Fovea from Colour Fundus Images

Lecture notes in networks and systems, Journal Year: 2024, Volume and Issue: unknown, P. 383 - 395

Published: Jan. 1, 2024

Language: Английский

---

Enhancing Retina Images by Lowpass Filtering Using Binomial Filter

Diagnostics, Journal Year: 2024, Volume and Issue: 14(15), P. 1688 - 1688

Published: Aug. 5, 2024

This study presents a method to enhance the contrast and luminosity of fundus images with boundary reflection. In this work, 100 retina taken from online databases are utilized test performance proposed method. First, red, green blue channels read stored in separate arrays. Then, area eye also called region interest (ROI) is located by thresholding. Next, ratios R G B at every pixel ROI calculated along copies R, channels. RGB subjected average filtering using 3 × mask smoothen values pixels, especially border ROI. background brightness estimation stage, three filtered binomial filters (BFs). step creates (BB) surface levelling foreground objects like blood vessels, fundi, optic discs spots, thus allowing illumination. next BB, equalized so that all pixels will have same brightness. followed adjustment CLAHE. Afterward, details adjusted channel enhanced information red color correction intensities according their original before reunited. The resulting image resembles one distribution tone but shows marked improvement contrast. effectiveness approach tested on enhancement noticeable visually quantitatively greyscale color. On average, manages increase images. was implemented MATLAB R2021b an AMD 5900HS processor execution time less than 10 s. filter compared those two other it better results. technique can be useful tool for ophthalmologists who perform diagnoses eyes diabetic patients.

Language: Английский

---

Hard Exudates Segmentation in Diabetic retinopathy using DiaRetDB1

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 126486 - 126502

Published: Jan. 1, 2024

Language: Английский

---