Iterative nick thresholding region growing for hemorrhage and exudate detection and segmentation to prescreen diabetic retinopathy

Name: Patsaphon Chandhakanond

keyword: DR detection

; Hemorrhage segmentation

; Exudate segmentation

; Iterative nick thresholding region growing (INRG)

; Watershed separation (WS)

Abstract: Diabetic retinopathy (DR) is a prevalent eye condition affecting approximately one-third of individuals with diabetes, leading to vision loss in both working-age adults and the elderly. Early detection and intervention are crucial for improving patient outcomes and reducing the strain on healthcare systems. Developing advanced computational techniques enables the creation of automated systems for screening and managing DR. This study focuses on detecting and segmenting exudates and hemorrhages in retinal fundus images. We utilized the iterative nick thresholding region growing (INRG) method as the foundation of our approach. To enhance performance across different applications, we integrated the watershed separation (WS) algorithm and the Chi2 feature selection method into expanded feature sets. These enhancements were combined with the INRG method to effectively segment hemorrhages and exudates. The segmented results were then used to identify these features and ultimately detect diabetic retinopathy. Our method was evaluated by comparing its performance against two traditional approaches and two state-of-the-art techniques, including the original INRG-HSV model. For hemorrhage segmentation, the INRG method combined with WS (INRG-WS) achieved the best F-measure of 64.76%, surpassing all other methods. For exudate segmentation, the INRG-WS-Chi2 model, which combined the INRG method with WS and Chi2 ranking on expanded feature sets, delivered superior results. In hemorrhage detection, the INRG method without WS but utilizing hue, saturation, and brightness (INRG-HSV) achieved the highest accuracy of 90.27% with the lowest false negative rate (FNR) of 9.39%. For exudate detection, the INRG-WS-HSV model, which combined the INRG method with WS and HSV features, demonstrated the best accuracy of 88.14% and the lowest FNR of 8.75%. For diabetic retinopathy detection, the best-performing hemorrhage (INRG-HSV) and exudate (INRG-WS-HSV) models were compared against a state-of-the-art approach. Our models outperformed the benchmark, achieving an impressive accuracy of 89.89% and an FNR of 3.66%.

Thammasat University. Thammasat University Library

Address: BANGKOK

Email: preserv@tu.ac.th

Name: Pakinee Aimmanee

Role: advisor

Created: 2024

Modified: 2568-05-15

Issued: 2025-04-21

วิทยานิพนธ์/Thesis

application/pdf

DOI: 10.14457/TU.the.2024.92

eng

DegreeName: Master of Science

Level: Master's Degree

Descipline: Engineering and Technology

Grantor: Thammasat University

©copyrights Thammasat University

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	Patsaphon.pdf	2 MB

ใช้เวลา

0.022677 วินาที